kernel/trace/trace_events.c

   1 /*
   2  * event tracer
   3  *
   4  * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
   5  *
   6  *  - Added format output of fields of the trace point.
   7  *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
   8  *
   9  */
  10
  11 #include <linux/workqueue.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/kthread.h>
  14 #include <linux/debugfs.h>
  15 #include <linux/uaccess.h>
  16 #include <linux/module.h>
  17 #include <linux/ctype.h>
  18 #include <linux/slab.h>
  19 #include <linux/delay.h>
  20
  21 #include <asm/setup.h>
  22
  23 #include "trace_output.h"
  24
  25 #undef TRACE_SYSTEM
  26 #define TRACE_SYSTEM "TRACE_SYSTEM"
  27
  28 DEFINE_MUTEX(event_mutex);
  29
  30 DEFINE_MUTEX(event_storage_mutex);
  31 EXPORT_SYMBOL_GPL(event_storage_mutex);
  32
  33 char event_storage[EVENT_STORAGE_SIZE];
  34 EXPORT_SYMBOL_GPL(event_storage);
  35
  36 LIST_HEAD(ftrace_events);
  37 static LIST_HEAD(ftrace_common_fields);
  38
  39 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
  40
  41 static struct kmem_cache *field_cachep;
  42 static struct kmem_cache *file_cachep;
  43
  44 #define SYSTEM_FL_FREE_NAME             (1 << 31)
  45
  46 static inline int system_refcount(struct event_subsystem *system)
  47 {
  48         return system->ref_count & ~SYSTEM_FL_FREE_NAME;
  49 }
  50
  51 static int system_refcount_inc(struct event_subsystem *system)
  52 {
  53         return (system->ref_count++) & ~SYSTEM_FL_FREE_NAME;
  54 }
  55
  56 static int system_refcount_dec(struct event_subsystem *system)
  57 {
  58         return (--system->ref_count) & ~SYSTEM_FL_FREE_NAME;
  59 }
  60
  61 /* Double loops, do not use break, only goto's work */
  62 #define do_for_each_event_file(tr, file)                        \
  63         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  64                 list_for_each_entry(file, &tr->events, list)
  65
  66 #define do_for_each_event_file_safe(tr, file)                   \
  67         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  68                 struct ftrace_event_file *___n;                         \
  69                 list_for_each_entry_safe(file, ___n, &tr->events, list)
  70
  71 #define while_for_each_event_file()             \
  72         }
  73
  74 static struct list_head *
  75 trace_get_fields(struct ftrace_event_call *event_call)
  76 {
  77         if (!event_call->class->get_fields)
  78                 return &event_call->class->fields;
  79         return event_call->class->get_fields(event_call);
  80 }
  81
  82 static struct ftrace_event_field *
  83 __find_event_field(struct list_head *head, char *name)
  84 {
  85         struct ftrace_event_field *field;
  86
  87         list_for_each_entry(field, head, link) {
  88                 if (!strcmp(field->name, name))
  89                         return field;
  90         }
  91
  92         return NULL;
  93 }
  94
  95 struct ftrace_event_field *
  96 trace_find_event_field(struct ftrace_event_call *call, char *name)
  97 {
  98         struct ftrace_event_field *field;
  99         struct list_head *head;
 100
 101         field = __find_event_field(&ftrace_common_fields, name);
 102         if (field)
 103                 return field;
 104
 105         head = trace_get_fields(call);
 106         return __find_event_field(head, name);
 107 }
 108
 109 static int __trace_define_field(struct list_head *head, const char *type,
 110                                 const char *name, int offset, int size,
 111                                 int is_signed, int filter_type)
 112 {
 113         struct ftrace_event_field *field;
 114
 115         field = kmem_cache_alloc(field_cachep, GFP_TRACE);
 116         if (!field)
 117                 return -ENOMEM;
 118
 119         field->name = name;
 120         field->type = type;
 121
 122         if (filter_type == FILTER_OTHER)
 123                 field->filter_type = filter_assign_type(type);
 124         else
 125                 field->filter_type = filter_type;
 126
 127         field->offset = offset;
 128         field->size = size;
 129         field->is_signed = is_signed;
 130
 131         list_add(&field->link, head);
 132
 133         return 0;
 134 }
 135
 136 int trace_define_field(struct ftrace_event_call *call, const char *type,
 137                        const char *name, int offset, int size, int is_signed,
 138                        int filter_type)
 139 {
 140         struct list_head *head;
 141
 142         if (WARN_ON(!call->class))
 143                 return 0;
 144
 145         head = trace_get_fields(call);
 146         return __trace_define_field(head, type, name, offset, size,
 147                                     is_signed, filter_type);
 148 }
 149 EXPORT_SYMBOL_GPL(trace_define_field);
 150
 151 #define __common_field(type, item)                                      \
 152         ret = __trace_define_field(&ftrace_common_fields, #type,        \
 153                                    "common_" #item,                     \
 154                                    offsetof(typeof(ent), item),         \
 155                                    sizeof(ent.item),                    \
 156                                    is_signed_type(type), FILTER_OTHER); \
 157         if (ret)                                                        \
 158                 return ret;
 159
 160 static int trace_define_common_fields(void)
 161 {
 162         int ret;
 163         struct trace_entry ent;
 164
 165         __common_field(unsigned short, type);
 166         __common_field(unsigned char, flags);
 167         __common_field(unsigned char, preempt_count);
 168         __common_field(int, pid);
 169
 170         return ret;
 171 }
 172
 173 static void trace_destroy_fields(struct ftrace_event_call *call)
 174 {
 175         struct ftrace_event_field *field, *next;
 176         struct list_head *head;
 177
 178         head = trace_get_fields(call);
 179         list_for_each_entry_safe(field, next, head, link) {
 180                 list_del(&field->link);
 181                 kmem_cache_free(field_cachep, field);
 182         }
 183 }
 184
 185 int trace_event_raw_init(struct ftrace_event_call *call)
 186 {
 187         int id;
 188
 189         id = register_ftrace_event(&call->event);
 190         if (!id)
 191                 return -ENODEV;
 192
 193         return 0;
 194 }
 195 EXPORT_SYMBOL_GPL(trace_event_raw_init);
 196
 197 int ftrace_event_reg(struct ftrace_event_call *call,
 198                      enum trace_reg type, void *data)
 199 {
 200         struct ftrace_event_file *file = data;
 201
 202         switch (type) {
 203         case TRACE_REG_REGISTER:
 204                 return tracepoint_probe_register(call->name,
 205                                                  call->class->probe,
 206                                                  file);
 207         case TRACE_REG_UNREGISTER:
 208                 tracepoint_probe_unregister(call->name,
 209                                             call->class->probe,
 210                                             file);
 211                 return 0;
 212
 213 #ifdef CONFIG_PERF_EVENTS
 214         case TRACE_REG_PERF_REGISTER:
 215                 return tracepoint_probe_register(call->name,
 216                                                  call->class->perf_probe,
 217                                                  call);
 218         case TRACE_REG_PERF_UNREGISTER:
 219                 tracepoint_probe_unregister(call->name,
 220                                             call->class->perf_probe,
 221                                             call);
 222                 return 0;
 223         case TRACE_REG_PERF_OPEN:
 224         case TRACE_REG_PERF_CLOSE:
 225         case TRACE_REG_PERF_ADD:
 226         case TRACE_REG_PERF_DEL:
 227                 return 0;
 228 #endif
 229         }
 230         return 0;
 231 }
 232 EXPORT_SYMBOL_GPL(ftrace_event_reg);
 233
 234 void trace_event_enable_cmd_record(bool enable)
 235 {
 236         struct ftrace_event_file *file;
 237         struct trace_array *tr;
 238
 239         mutex_lock(&event_mutex);
 240         do_for_each_event_file(tr, file) {
 241
 242                 if (!(file->flags & FTRACE_EVENT_FL_ENABLED))
 243                         continue;
 244
 245                 if (enable) {
 246                         tracing_start_cmdline_record();
 247                         set_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
 248                 } else {
 249                         tracing_stop_cmdline_record();
 250                         clear_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
 251                 }
 252         } while_for_each_event_file();
 253         mutex_unlock(&event_mutex);
 254 }
 255
 256 static int __ftrace_event_enable_disable(struct ftrace_event_file *file,
 257                                          int enable, int soft_disable)
 258 {
 259         struct ftrace_event_call *call = file->event_call;
 260         int ret = 0;
 261         int disable;
 262
 263         switch (enable) {
 264         case 0:
 265                 /*
 266                  * When soft_disable is set and enable is cleared, the sm_ref
 267                  * reference counter is decremented. If it reaches 0, we want
 268                  * to clear the SOFT_DISABLED flag but leave the event in the
 269                  * state that it was. That is, if the event was enabled and
 270                  * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 271                  * is set we do not want the event to be enabled before we
 272                  * clear the bit.
 273                  *
 274                  * When soft_disable is not set but the SOFT_MODE flag is,
 275                  * we do nothing. Do not disable the tracepoint, otherwise
 276                  * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 277                  */
 278                 if (soft_disable) {
 279                         if (atomic_dec_return(&file->sm_ref) > 0)
 280                                 break;
 281                         disable = file->flags & FTRACE_EVENT_FL_SOFT_DISABLED;
 282                         clear_bit(FTRACE_EVENT_FL_SOFT_MODE_BIT, &file->flags);
 283                 } else
 284                         disable = !(file->flags & FTRACE_EVENT_FL_SOFT_MODE);
 285
 286                 if (disable && (file->flags & FTRACE_EVENT_FL_ENABLED)) {
 287                         clear_bit(FTRACE_EVENT_FL_ENABLED_BIT, &file->flags);
 288                         if (file->flags & FTRACE_EVENT_FL_RECORDED_CMD) {
 289                                 tracing_stop_cmdline_record();
 290                                 clear_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
 291                         }
 292                         call->class->reg(call, TRACE_REG_UNREGISTER, file);
 293                 }
 294                 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 295                 if (file->flags & FTRACE_EVENT_FL_SOFT_MODE)
 296                         set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
 297                 else
 298                         clear_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
 299                 break;
 300         case 1:
 301                 /*
 302                  * When soft_disable is set and enable is set, we want to
 303                  * register the tracepoint for the event, but leave the event
 304                  * as is. That means, if the event was already enabled, we do
 305                  * nothing (but set SOFT_MODE). If the event is disabled, we
 306                  * set SOFT_DISABLED before enabling the event tracepoint, so
 307                  * it still seems to be disabled.
 308                  */
 309                 if (!soft_disable)
 310                         clear_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
 311                 else {
 312                         if (atomic_inc_return(&file->sm_ref) > 1)
 313                                 break;
 314                         set_bit(FTRACE_EVENT_FL_SOFT_MODE_BIT, &file->flags);
 315                 }
 316
 317                 if (!(file->flags & FTRACE_EVENT_FL_ENABLED)) {
 318
 319                         /* Keep the event disabled, when going to SOFT_MODE. */
 320                         if (soft_disable)
 321                                 set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
 322
 323                         if (trace_flags & TRACE_ITER_RECORD_CMD) {
 324                                 tracing_start_cmdline_record();
 325                                 set_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
 326                         }
 327                         ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 328                         if (ret) {
 329                                 tracing_stop_cmdline_record();
 330                                 pr_info("event trace: Could not enable event "
 331                                         "%s\n", call->name);
 332                                 break;
 333                         }
 334                         set_bit(FTRACE_EVENT_FL_ENABLED_BIT, &file->flags);
 335
 336                         /* WAS_ENABLED gets set but never cleared. */
 337                         call->flags |= TRACE_EVENT_FL_WAS_ENABLED;
 338                 }
 339                 break;
 340         }
 341
 342         return ret;
 343 }
 344
 345 static int ftrace_event_enable_disable(struct ftrace_event_file *file,
 346                                        int enable)
 347 {
 348         return __ftrace_event_enable_disable(file, enable, 0);
 349 }
 350
 351 static void ftrace_clear_events(struct trace_array *tr)
 352 {
 353         struct ftrace_event_file *file;
 354
 355         mutex_lock(&event_mutex);
 356         list_for_each_entry(file, &tr->events, list) {
 357                 ftrace_event_enable_disable(file, 0);
 358         }
 359         mutex_unlock(&event_mutex);
 360 }
 361
 362 static void __put_system(struct event_subsystem *system)
 363 {
 364         struct event_filter *filter = system->filter;
 365
 366         WARN_ON_ONCE(system_refcount(system) == 0);
 367         if (system_refcount_dec(system))
 368                 return;
 369
 370         list_del(&system->list);
 371
 372         if (filter) {
 373                 kfree(filter->filter_string);
 374                 kfree(filter);
 375         }
 376         if (system->ref_count & SYSTEM_FL_FREE_NAME)
 377                 kfree(system->name);
 378         kfree(system);
 379 }
 380
 381 static void __get_system(struct event_subsystem *system)
 382 {
 383         WARN_ON_ONCE(system_refcount(system) == 0);
 384         system_refcount_inc(system);
 385 }
 386
 387 static void __get_system_dir(struct ftrace_subsystem_dir *dir)
 388 {
 389         WARN_ON_ONCE(dir->ref_count == 0);
 390         dir->ref_count++;
 391         __get_system(dir->subsystem);
 392 }
 393
 394 static void __put_system_dir(struct ftrace_subsystem_dir *dir)
 395 {
 396         WARN_ON_ONCE(dir->ref_count == 0);
 397         /* If the subsystem is about to be freed, the dir must be too */
 398         WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
 399
 400         __put_system(dir->subsystem);
 401         if (!--dir->ref_count)
 402                 kfree(dir);
 403 }
 404
 405 static void put_system(struct ftrace_subsystem_dir *dir)
 406 {
 407         mutex_lock(&event_mutex);
 408         __put_system_dir(dir);
 409         mutex_unlock(&event_mutex);
 410 }
 411
 412 /*
 413  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 414  */
 415 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
 416                                   const char *sub, const char *event, int set)
 417 {
 418         struct ftrace_event_file *file;
 419         struct ftrace_event_call *call;
 420         int ret = -EINVAL;
 421
 422         mutex_lock(&event_mutex);
 423         list_for_each_entry(file, &tr->events, list) {
 424
 425                 call = file->event_call;
 426
 427                 if (!call->name || !call->class || !call->class->reg)
 428                         continue;
 429
 430                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
 431                         continue;
 432
 433                 if (match &&
 434                     strcmp(match, call->name) != 0 &&
 435                     strcmp(match, call->class->system) != 0)
 436                         continue;
 437
 438                 if (sub && strcmp(sub, call->class->system) != 0)
 439                         continue;
 440
 441                 if (event && strcmp(event, call->name) != 0)
 442                         continue;
 443
 444                 ftrace_event_enable_disable(file, set);
 445
 446                 ret = 0;
 447         }
 448         mutex_unlock(&event_mutex);
 449
 450         return ret;
 451 }
 452
 453 static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
 454 {
 455         char *event = NULL, *sub = NULL, *match;
 456
 457         /*
 458          * The buf format can be <subsystem>:<event-name>
 459          *  *:<event-name> means any event by that name.
 460          *  :<event-name> is the same.
 461          *
 462          *  <subsystem>:* means all events in that subsystem
 463          *  <subsystem>: means the same.
 464          *
 465          *  <name> (no ':') means all events in a subsystem with
 466          *  the name <name> or any event that matches <name>
 467          */
 468
 469         match = strsep(&buf, ":");
 470         if (buf) {
 471                 sub = match;
 472                 event = buf;
 473                 match = NULL;
 474
 475                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 476                         sub = NULL;
 477                 if (!strlen(event) || strcmp(event, "*") == 0)
 478                         event = NULL;
 479         }
 480
 481         return __ftrace_set_clr_event(tr, match, sub, event, set);
 482 }
 483
 484 /**
 485  * trace_set_clr_event - enable or disable an event
 486  * @system: system name to match (NULL for any system)
 487  * @event: event name to match (NULL for all events, within system)
 488  * @set: 1 to enable, 0 to disable
 489  *
 490  * This is a way for other parts of the kernel to enable or disable
 491  * event recording.
 492  *
 493  * Returns 0 on success, -EINVAL if the parameters do not match any
 494  * registered events.
 495  */
 496 int trace_set_clr_event(const char *system, const char *event, int set)
 497 {
 498         struct trace_array *tr = top_trace_array();
 499
 500         return __ftrace_set_clr_event(tr, NULL, system, event, set);
 501 }
 502 EXPORT_SYMBOL_GPL(trace_set_clr_event);
 503
 504 /* 128 should be much more than enough */
 505 #define EVENT_BUF_SIZE          127
 506
 507 static ssize_t
 508 ftrace_event_write(struct file *file, const char __user *ubuf,
 509                    size_t cnt, loff_t *ppos)
 510 {
 511         struct trace_parser parser;
 512         struct seq_file *m = file->private_data;
 513         struct trace_array *tr = m->private;
 514         ssize_t read, ret;
 515
 516         if (!cnt)
 517                 return 0;
 518
 519         ret = tracing_update_buffers();
 520         if (ret < 0)
 521                 return ret;
 522
 523         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 524                 return -ENOMEM;
 525
 526         read = trace_get_user(&parser, ubuf, cnt, ppos);
 527
 528         if (read >= 0 && trace_parser_loaded((&parser))) {
 529                 int set = 1;
 530
 531                 if (*parser.buffer == '!')
 532                         set = 0;
 533
 534                 parser.buffer[parser.idx] = 0;
 535
 536                 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
 537                 if (ret)
 538                         goto out_put;
 539         }
 540
 541         ret = read;
 542
 543  out_put:
 544         trace_parser_put(&parser);
 545
 546         return ret;
 547 }
 548
 549 static void *
 550 t_next(struct seq_file *m, void *v, loff_t *pos)
 551 {
 552         struct ftrace_event_file *file = v;
 553         struct ftrace_event_call *call;
 554         struct trace_array *tr = m->private;
 555
 556         (*pos)++;
 557
 558         list_for_each_entry_continue(file, &tr->events, list) {
 559                 call = file->event_call;
 560                 /*
 561                  * The ftrace subsystem is for showing formats only.
 562                  * They can not be enabled or disabled via the event files.
 563                  */
 564                 if (call->class && call->class->reg)
 565                         return file;
 566         }
 567
 568         return NULL;
 569 }
 570
 571 static void *t_start(struct seq_file *m, loff_t *pos)
 572 {
 573         struct ftrace_event_file *file;
 574         struct trace_array *tr = m->private;
 575         loff_t l;
 576
 577         mutex_lock(&event_mutex);
 578
 579         file = list_entry(&tr->events, struct ftrace_event_file, list);
 580         for (l = 0; l <= *pos; ) {
 581                 file = t_next(m, file, &l);
 582                 if (!file)
 583                         break;
 584         }
 585         return file;
 586 }
 587
 588 static void *
 589 s_next(struct seq_file *m, void *v, loff_t *pos)
 590 {
 591         struct ftrace_event_file *file = v;
 592         struct trace_array *tr = m->private;
 593
 594         (*pos)++;
 595
 596         list_for_each_entry_continue(file, &tr->events, list) {
 597                 if (file->flags & FTRACE_EVENT_FL_ENABLED)
 598                         return file;
 599         }
 600
 601         return NULL;
 602 }
 603
 604 static void *s_start(struct seq_file *m, loff_t *pos)
 605 {
 606         struct ftrace_event_file *file;
 607         struct trace_array *tr = m->private;
 608         loff_t l;
 609
 610         mutex_lock(&event_mutex);
 611
 612         file = list_entry(&tr->events, struct ftrace_event_file, list);
 613         for (l = 0; l <= *pos; ) {
 614                 file = s_next(m, file, &l);
 615                 if (!file)
 616                         break;
 617         }
 618         return file;
 619 }
 620
 621 static int t_show(struct seq_file *m, void *v)
 622 {
 623         struct ftrace_event_file *file = v;
 624         struct ftrace_event_call *call = file->event_call;
 625
 626         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
 627                 seq_printf(m, "%s:", call->class->system);
 628         seq_printf(m, "%s\n", call->name);
 629
 630         return 0;
 631 }
 632
 633 static void t_stop(struct seq_file *m, void *p)
 634 {
 635         mutex_unlock(&event_mutex);
 636 }
 637
 638 static ssize_t
 639 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 640                   loff_t *ppos)
 641 {
 642         struct ftrace_event_file *file = filp->private_data;
 643         char buf[4] = "0";
 644
 645         if (file->flags & FTRACE_EVENT_FL_ENABLED &&
 646             !(file->flags & FTRACE_EVENT_FL_SOFT_DISABLED))
 647                 strcpy(buf, "1");
 648
 649         if (file->flags & FTRACE_EVENT_FL_SOFT_DISABLED ||
 650             file->flags & FTRACE_EVENT_FL_SOFT_MODE)
 651                 strcat(buf, "*");
 652
 653         strcat(buf, "\n");
 654
 655         return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
 656 }
 657
 658 static ssize_t
 659 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 660                    loff_t *ppos)
 661 {
 662         struct ftrace_event_file *file = filp->private_data;
 663         unsigned long val;
 664         int ret;
 665
 666         if (!file)
 667                 return -EINVAL;
 668
 669         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
 670         if (ret)
 671                 return ret;
 672
 673         ret = tracing_update_buffers();
 674         if (ret < 0)
 675                 return ret;
 676
 677         switch (val) {
 678         case 0:
 679         case 1:
 680                 mutex_lock(&event_mutex);
 681                 ret = ftrace_event_enable_disable(file, val);
 682                 mutex_unlock(&event_mutex);
 683                 break;
 684
 685         default:
 686                 return -EINVAL;
 687         }
 688
 689         *ppos += cnt;
 690
 691         return ret ? ret : cnt;
 692 }
 693
 694 static ssize_t
 695 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 696                    loff_t *ppos)
 697 {
 698         const char set_to_char[4] = { '?', '0', '1', 'X' };
 699         struct ftrace_subsystem_dir *dir = filp->private_data;
 700         struct event_subsystem *system = dir->subsystem;
 701         struct ftrace_event_call *call;
 702         struct ftrace_event_file *file;
 703         struct trace_array *tr = dir->tr;
 704         char buf[2];
 705         int set = 0;
 706         int ret;
 707
 708         mutex_lock(&event_mutex);
 709         list_for_each_entry(file, &tr->events, list) {
 710                 call = file->event_call;
 711                 if (!call->name || !call->class || !call->class->reg)
 712                         continue;
 713
 714                 if (system && strcmp(call->class->system, system->name) != 0)
 715                         continue;
 716
 717                 /*
 718                  * We need to find out if all the events are set
 719                  * or if all events or cleared, or if we have
 720                  * a mixture.
 721                  */
 722                 set |= (1 << !!(file->flags & FTRACE_EVENT_FL_ENABLED));
 723
 724                 /*
 725                  * If we have a mixture, no need to look further.
 726                  */
 727                 if (set == 3)
 728                         break;
 729         }
 730         mutex_unlock(&event_mutex);
 731
 732         buf[0] = set_to_char[set];
 733         buf[1] = '\n';
 734
 735         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 736
 737         return ret;
 738 }
 739
 740 static ssize_t
 741 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 742                     loff_t *ppos)
 743 {
 744         struct ftrace_subsystem_dir *dir = filp->private_data;
 745         struct event_subsystem *system = dir->subsystem;
 746         const char *name = NULL;
 747         unsigned long val;
 748         ssize_t ret;
 749
 750         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
 751         if (ret)
 752                 return ret;
 753
 754         ret = tracing_update_buffers();
 755         if (ret < 0)
 756                 return ret;
 757
 758         if (val != 0 && val != 1)
 759                 return -EINVAL;
 760
 761         /*
 762          * Opening of "enable" adds a ref count to system,
 763          * so the name is safe to use.
 764          */
 765         if (system)
 766                 name = system->name;
 767
 768         ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
 769         if (ret)
 770                 goto out;
 771
 772         ret = cnt;
 773
 774 out:
 775         *ppos += cnt;
 776
 777         return ret;
 778 }
 779
 780 enum {
 781         FORMAT_HEADER           = 1,
 782         FORMAT_FIELD_SEPERATOR  = 2,
 783         FORMAT_PRINTFMT         = 3,
 784 };
 785
 786 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
 787 {
 788         struct ftrace_event_call *call = m->private;
 789         struct ftrace_event_field *field;
 790         struct list_head *common_head = &ftrace_common_fields;
 791         struct list_head *head = trace_get_fields(call);
 792
 793         (*pos)++;
 794
 795         switch ((unsigned long)v) {
 796         case FORMAT_HEADER:
 797                 if (unlikely(list_empty(common_head)))
 798                         return NULL;
 799
 800                 field = list_entry(common_head->prev,
 801                                    struct ftrace_event_field, link);
 802                 return field;
 803
 804         case FORMAT_FIELD_SEPERATOR:
 805                 if (unlikely(list_empty(head)))
 806                         return NULL;
 807
 808                 field = list_entry(head->prev, struct ftrace_event_field, link);
 809                 return field;
 810
 811         case FORMAT_PRINTFMT:
 812                 /* all done */
 813                 return NULL;
 814         }
 815
 816         field = v;
 817         if (field->link.prev == common_head)
 818                 return (void *)FORMAT_FIELD_SEPERATOR;
 819         else if (field->link.prev == head)
 820                 return (void *)FORMAT_PRINTFMT;
 821
 822         field = list_entry(field->link.prev, struct ftrace_event_field, link);
 823
 824         return field;
 825 }
 826
 827 static void *f_start(struct seq_file *m, loff_t *pos)
 828 {
 829         loff_t l = 0;
 830         void *p;
 831
 832         /* Start by showing the header */
 833         if (!*pos)
 834                 return (void *)FORMAT_HEADER;
 835
 836         p = (void *)FORMAT_HEADER;
 837         do {
 838                 p = f_next(m, p, &l);
 839         } while (p && l < *pos);
 840
 841         return p;
 842 }
 843
 844 static int f_show(struct seq_file *m, void *v)
 845 {
 846         struct ftrace_event_call *call = m->private;
 847         struct ftrace_event_field *field;
 848         const char *array_descriptor;
 849
 850         switch ((unsigned long)v) {
 851         case FORMAT_HEADER:
 852                 seq_printf(m, "name: %s\n", call->name);
 853                 seq_printf(m, "ID: %d\n", call->event.type);
 854                 seq_printf(m, "format:\n");
 855                 return 0;
 856
 857         case FORMAT_FIELD_SEPERATOR:
 858                 seq_putc(m, '\n');
 859                 return 0;
 860
 861         case FORMAT_PRINTFMT:
 862                 seq_printf(m, "\nprint fmt: %s\n",
 863                            call->print_fmt);
 864                 return 0;
 865         }
 866
 867         field = v;
 868
 869         /*
 870          * Smartly shows the array type(except dynamic array).
 871          * Normal:
 872          *      field:TYPE VAR
 873          * If TYPE := TYPE[LEN], it is shown:
 874          *      field:TYPE VAR[LEN]
 875          */
 876         array_descriptor = strchr(field->type, '[');
 877
 878         if (!strncmp(field->type, "__data_loc", 10))
 879                 array_descriptor = NULL;
 880
 881         if (!array_descriptor)
 882                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
 883                            field->type, field->name, field->offset,
 884                            field->size, !!field->is_signed);
 885         else
 886                 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
 887                            (int)(array_descriptor - field->type),
 888                            field->type, field->name,
 889                            array_descriptor, field->offset,
 890                            field->size, !!field->is_signed);
 891
 892         return 0;
 893 }
 894
 895 static void f_stop(struct seq_file *m, void *p)
 896 {
 897 }
 898
 899 static const struct seq_operations trace_format_seq_ops = {
 900         .start          = f_start,
 901         .next           = f_next,
 902         .stop           = f_stop,
 903         .show           = f_show,
 904 };
 905
 906 static int trace_format_open(struct inode *inode, struct file *file)
 907 {
 908         struct ftrace_event_call *call = inode->i_private;
 909         struct seq_file *m;
 910         int ret;
 911
 912         ret = seq_open(file, &trace_format_seq_ops);
 913         if (ret < 0)
 914                 return ret;
 915
 916         m = file->private_data;
 917         m->private = call;
 918
 919         return 0;
 920 }
 921
 922 static ssize_t
 923 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 924 {
 925         struct ftrace_event_call *call = filp->private_data;
 926         struct trace_seq *s;
 927         int r;
 928
 929         if (*ppos)
 930                 return 0;
 931
 932         s = kmalloc(sizeof(*s), GFP_KERNEL);
 933         if (!s)
 934                 return -ENOMEM;
 935
 936         trace_seq_init(s);
 937         trace_seq_printf(s, "%d\n", call->event.type);
 938
 939         r = simple_read_from_buffer(ubuf, cnt, ppos,
 940                                     s->buffer, s->len);
 941         kfree(s);
 942         return r;
 943 }
 944
 945 static ssize_t
 946 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 947                   loff_t *ppos)
 948 {
 949         struct ftrace_event_call *call = filp->private_data;
 950         struct trace_seq *s;
 951         int r;
 952
 953         if (*ppos)
 954                 return 0;
 955
 956         s = kmalloc(sizeof(*s), GFP_KERNEL);
 957         if (!s)
 958                 return -ENOMEM;
 959
 960         trace_seq_init(s);
 961
 962         print_event_filter(call, s);
 963         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 964
 965         kfree(s);
 966
 967         return r;
 968 }
 969
 970 static ssize_t
 971 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
 972                    loff_t *ppos)
 973 {
 974         struct ftrace_event_call *call = filp->private_data;
 975         char *buf;
 976         int err;
 977
 978         if (cnt >= PAGE_SIZE)
 979                 return -EINVAL;
 980
 981         buf = (char *)__get_free_page(GFP_TEMPORARY);
 982         if (!buf)
 983                 return -ENOMEM;
 984
 985         if (copy_from_user(buf, ubuf, cnt)) {
 986                 free_page((unsigned long) buf);
 987                 return -EFAULT;
 988         }
 989         buf[cnt] = '\0';
 990
 991         err = apply_event_filter(call, buf);
 992         free_page((unsigned long) buf);
 993         if (err < 0)
 994                 return err;
 995
 996         *ppos += cnt;
 997
 998         return cnt;
 999 }
1000
1001 static LIST_HEAD(event_subsystems);
1002
1003 static int subsystem_open(struct inode *inode, struct file *filp)
1004 {
1005         struct event_subsystem *system = NULL;
1006         struct ftrace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1007         struct trace_array *tr;
1008         int ret;
1009
1010         /* Make sure the system still exists */
1011         mutex_lock(&trace_types_lock);
1012         mutex_lock(&event_mutex);
1013         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1014                 list_for_each_entry(dir, &tr->systems, list) {
1015                         if (dir == inode->i_private) {
1016                                 /* Don't open systems with no events */
1017                                 if (dir->nr_events) {
1018                                         __get_system_dir(dir);
1019                                         system = dir->subsystem;
1020                                 }
1021                                 goto exit_loop;
1022                         }
1023                 }
1024         }
1025  exit_loop:
1026         mutex_unlock(&event_mutex);
1027         mutex_unlock(&trace_types_lock);
1028
1029         if (!system)
1030                 return -ENODEV;
1031
1032         /* Some versions of gcc think dir can be uninitialized here */
1033         WARN_ON(!dir);
1034
1035         ret = tracing_open_generic(inode, filp);
1036         if (ret < 0)
1037                 put_system(dir);
1038
1039         return ret;
1040 }
1041
1042 static int system_tr_open(struct inode *inode, struct file *filp)
1043 {
1044         struct ftrace_subsystem_dir *dir;
1045         struct trace_array *tr = inode->i_private;
1046         int ret;
1047
1048         /* Make a temporary dir that has no system but points to tr */
1049         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1050         if (!dir)
1051                 return -ENOMEM;
1052
1053         dir->tr = tr;
1054
1055         ret = tracing_open_generic(inode, filp);
1056         if (ret < 0)
1057                 kfree(dir);
1058
1059         filp->private_data = dir;
1060
1061         return ret;
1062 }
1063
1064 static int subsystem_release(struct inode *inode, struct file *file)
1065 {
1066         struct ftrace_subsystem_dir *dir = file->private_data;
1067
1068         /*
1069          * If dir->subsystem is NULL, then this is a temporary
1070          * descriptor that was made for a trace_array to enable
1071          * all subsystems.
1072          */
1073         if (dir->subsystem)
1074                 put_system(dir);
1075         else
1076                 kfree(dir);
1077
1078         return 0;
1079 }
1080
1081 static ssize_t
1082 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1083                       loff_t *ppos)
1084 {
1085         struct ftrace_subsystem_dir *dir = filp->private_data;
1086         struct event_subsystem *system = dir->subsystem;
1087         struct trace_seq *s;
1088         int r;
1089
1090         if (*ppos)
1091                 return 0;
1092
1093         s = kmalloc(sizeof(*s), GFP_KERNEL);
1094         if (!s)
1095                 return -ENOMEM;
1096
1097         trace_seq_init(s);
1098
1099         print_subsystem_event_filter(system, s);
1100         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
1101
1102         kfree(s);
1103
1104         return r;
1105 }
1106
1107 static ssize_t
1108 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1109                        loff_t *ppos)
1110 {
1111         struct ftrace_subsystem_dir *dir = filp->private_data;
1112         char *buf;
1113         int err;
1114
1115         if (cnt >= PAGE_SIZE)
1116                 return -EINVAL;
1117
1118         buf = (char *)__get_free_page(GFP_TEMPORARY);
1119         if (!buf)
1120                 return -ENOMEM;
1121
1122         if (copy_from_user(buf, ubuf, cnt)) {
1123                 free_page((unsigned long) buf);
1124                 return -EFAULT;
1125         }
1126         buf[cnt] = '\0';
1127
1128         err = apply_subsystem_event_filter(dir, buf);
1129         free_page((unsigned long) buf);
1130         if (err < 0)
1131                 return err;
1132
1133         *ppos += cnt;
1134
1135         return cnt;
1136 }
1137
1138 static ssize_t
1139 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1140 {
1141         int (*func)(struct trace_seq *s) = filp->private_data;
1142         struct trace_seq *s;
1143         int r;
1144
1145         if (*ppos)
1146                 return 0;
1147
1148         s = kmalloc(sizeof(*s), GFP_KERNEL);
1149         if (!s)
1150                 return -ENOMEM;
1151
1152         trace_seq_init(s);
1153
1154         func(s);
1155         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
1156
1157         kfree(s);
1158
1159         return r;
1160 }
1161
1162 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1163 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1164
1165 static const struct seq_operations show_event_seq_ops = {
1166         .start = t_start,
1167         .next = t_next,
1168         .show = t_show,
1169         .stop = t_stop,
1170 };
1171
1172 static const struct seq_operations show_set_event_seq_ops = {
1173         .start = s_start,
1174         .next = s_next,
1175         .show = t_show,
1176         .stop = t_stop,
1177 };
1178
1179 static const struct file_operations ftrace_avail_fops = {
1180         .open = ftrace_event_avail_open,
1181         .read = seq_read,
1182         .llseek = seq_lseek,
1183         .release = seq_release,
1184 };
1185
1186 static const struct file_operations ftrace_set_event_fops = {
1187         .open = ftrace_event_set_open,
1188         .read = seq_read,
1189         .write = ftrace_event_write,
1190         .llseek = seq_lseek,
1191         .release = seq_release,
1192 };
1193
1194 static const struct file_operations ftrace_enable_fops = {
1195         .open = tracing_open_generic,
1196         .read = event_enable_read,
1197         .write = event_enable_write,
1198         .llseek = default_llseek,
1199 };
1200
1201 static const struct file_operations ftrace_event_format_fops = {
1202         .open = trace_format_open,
1203         .read = seq_read,
1204         .llseek = seq_lseek,
1205         .release = seq_release,
1206 };
1207
1208 static const struct file_operations ftrace_event_id_fops = {
1209         .open = tracing_open_generic,
1210         .read = event_id_read,
1211         .llseek = default_llseek,
1212 };
1213
1214 static const struct file_operations ftrace_event_filter_fops = {
1215         .open = tracing_open_generic,
1216         .read = event_filter_read,
1217         .write = event_filter_write,
1218         .llseek = default_llseek,
1219 };
1220
1221 static const struct file_operations ftrace_subsystem_filter_fops = {
1222         .open = subsystem_open,
1223         .read = subsystem_filter_read,
1224         .write = subsystem_filter_write,
1225         .llseek = default_llseek,
1226         .release = subsystem_release,
1227 };
1228
1229 static const struct file_operations ftrace_system_enable_fops = {
1230         .open = subsystem_open,
1231         .read = system_enable_read,
1232         .write = system_enable_write,
1233         .llseek = default_llseek,
1234         .release = subsystem_release,
1235 };
1236
1237 static const struct file_operations ftrace_tr_enable_fops = {
1238         .open = system_tr_open,
1239         .read = system_enable_read,
1240         .write = system_enable_write,
1241         .llseek = default_llseek,
1242         .release = subsystem_release,
1243 };
1244
1245 static const struct file_operations ftrace_show_header_fops = {
1246         .open = tracing_open_generic,
1247         .read = show_header,
1248         .llseek = default_llseek,
1249 };
1250
1251 static int
1252 ftrace_event_open(struct inode *inode, struct file *file,
1253                   const struct seq_operations *seq_ops)
1254 {
1255         struct seq_file *m;
1256         int ret;
1257
1258         ret = seq_open(file, seq_ops);
1259         if (ret < 0)
1260                 return ret;
1261         m = file->private_data;
1262         /* copy tr over to seq ops */
1263         m->private = inode->i_private;
1264
1265         return ret;
1266 }
1267
1268 static int
1269 ftrace_event_avail_open(struct inode *inode, struct file *file)
1270 {
1271         const struct seq_operations *seq_ops = &show_event_seq_ops;
1272
1273         return ftrace_event_open(inode, file, seq_ops);
1274 }
1275
1276 static int
1277 ftrace_event_set_open(struct inode *inode, struct file *file)
1278 {
1279         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1280         struct trace_array *tr = inode->i_private;
1281
1282         if ((file->f_mode & FMODE_WRITE) &&
1283             (file->f_flags & O_TRUNC))
1284                 ftrace_clear_events(tr);
1285
1286         return ftrace_event_open(inode, file, seq_ops);
1287 }
1288
1289 static struct event_subsystem *
1290 create_new_subsystem(const char *name)
1291 {
1292         struct event_subsystem *system;
1293
1294         /* need to create new entry */
1295         system = kmalloc(sizeof(*system), GFP_KERNEL);
1296         if (!system)
1297                 return NULL;
1298
1299         system->ref_count = 1;
1300
1301         /* Only allocate if dynamic (kprobes and modules) */
1302         if (!core_kernel_data((unsigned long)name)) {
1303                 system->ref_count |= SYSTEM_FL_FREE_NAME;
1304                 system->name = kstrdup(name, GFP_KERNEL);
1305                 if (!system->name)
1306                         goto out_free;
1307         } else
1308                 system->name = name;
1309
1310         system->filter = NULL;
1311
1312         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1313         if (!system->filter)
1314                 goto out_free;
1315
1316         list_add(&system->list, &event_subsystems);
1317
1318         return system;
1319
1320  out_free:
1321         if (system->ref_count & SYSTEM_FL_FREE_NAME)
1322                 kfree(system->name);
1323         kfree(system);
1324         return NULL;
1325 }
1326
1327 static struct dentry *
1328 event_subsystem_dir(struct trace_array *tr, const char *name,
1329                     struct ftrace_event_file *file, struct dentry *parent)
1330 {
1331         struct ftrace_subsystem_dir *dir;
1332         struct event_subsystem *system;
1333         struct dentry *entry;
1334
1335         /* First see if we did not already create this dir */
1336         list_for_each_entry(dir, &tr->systems, list) {
1337                 system = dir->subsystem;
1338                 if (strcmp(system->name, name) == 0) {
1339                         dir->nr_events++;
1340                         file->system = dir;
1341                         return dir->entry;
1342                 }
1343         }
1344
1345         /* Now see if the system itself exists. */
1346         list_for_each_entry(system, &event_subsystems, list) {
1347                 if (strcmp(system->name, name) == 0)
1348                         break;
1349         }
1350         /* Reset system variable when not found */
1351         if (&system->list == &event_subsystems)
1352                 system = NULL;
1353
1354         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1355         if (!dir)
1356                 goto out_fail;
1357
1358         if (!system) {
1359                 system = create_new_subsystem(name);
1360                 if (!system)
1361                         goto out_free;
1362         } else
1363                 __get_system(system);
1364
1365         dir->entry = debugfs_create_dir(name, parent);
1366         if (!dir->entry) {
1367                 pr_warning("Failed to create system directory %s\n", name);
1368                 __put_system(system);
1369                 goto out_free;
1370         }
1371
1372         dir->tr = tr;
1373         dir->ref_count = 1;
1374         dir->nr_events = 1;
1375         dir->subsystem = system;
1376         file->system = dir;
1377
1378         entry = debugfs_create_file("filter", 0644, dir->entry, dir,
1379                                     &ftrace_subsystem_filter_fops);
1380         if (!entry) {
1381                 kfree(system->filter);
1382                 system->filter = NULL;
1383                 pr_warning("Could not create debugfs '%s/filter' entry\n", name);
1384         }
1385
1386         trace_create_file("enable", 0644, dir->entry, dir,
1387                           &ftrace_system_enable_fops);
1388
1389         list_add(&dir->list, &tr->systems);
1390
1391         return dir->entry;
1392
1393  out_free:
1394         kfree(dir);
1395  out_fail:
1396         /* Only print this message if failed on memory allocation */
1397         if (!dir || !system)
1398                 pr_warning("No memory to create event subsystem %s\n",
1399                            name);
1400         return NULL;
1401 }
1402
1403 static int
1404 event_create_dir(struct dentry *parent,
1405                  struct ftrace_event_file *file,
1406                  const struct file_operations *id,
1407                  const struct file_operations *enable,
1408                  const struct file_operations *filter,
1409                  const struct file_operations *format)
1410 {
1411         struct ftrace_event_call *call = file->event_call;
1412         struct trace_array *tr = file->tr;
1413         struct list_head *head;
1414         struct dentry *d_events;
1415         int ret;
1416
1417         /*
1418          * If the trace point header did not define TRACE_SYSTEM
1419          * then the system would be called "TRACE_SYSTEM".
1420          */
1421         if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1422                 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1423                 if (!d_events)
1424                         return -ENOMEM;
1425         } else
1426                 d_events = parent;
1427
1428         file->dir = debugfs_create_dir(call->name, d_events);
1429         if (!file->dir) {
1430                 pr_warning("Could not create debugfs '%s' directory\n",
1431                            call->name);
1432                 return -1;
1433         }
1434
1435         if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1436                 trace_create_file("enable", 0644, file->dir, file,
1437                                   enable);
1438
1439 #ifdef CONFIG_PERF_EVENTS
1440         if (call->event.type && call->class->reg)
1441                 trace_create_file("id", 0444, file->dir, call,
1442                                   id);
1443 #endif
1444
1445         /*
1446          * Other events may have the same class. Only update
1447          * the fields if they are not already defined.
1448          */
1449         head = trace_get_fields(call);
1450         if (list_empty(head)) {
1451                 ret = call->class->define_fields(call);
1452                 if (ret < 0) {
1453                         pr_warning("Could not initialize trace point"
1454                                    " events/%s\n", call->name);
1455                         return -1;
1456                 }
1457         }
1458         trace_create_file("filter", 0644, file->dir, call,
1459                           filter);
1460
1461         trace_create_file("format", 0444, file->dir, call,
1462                           format);
1463
1464         return 0;
1465 }
1466
1467 static void remove_subsystem(struct ftrace_subsystem_dir *dir)
1468 {
1469         if (!dir)
1470                 return;
1471
1472         if (!--dir->nr_events) {
1473                 debugfs_remove_recursive(dir->entry);
1474                 list_del(&dir->list);
1475                 __put_system_dir(dir);
1476         }
1477 }
1478
1479 static void remove_event_from_tracers(struct ftrace_event_call *call)
1480 {
1481         struct ftrace_event_file *file;
1482         struct trace_array *tr;
1483
1484         do_for_each_event_file_safe(tr, file) {
1485
1486                 if (file->event_call != call)
1487                         continue;
1488
1489                 list_del(&file->list);
1490                 debugfs_remove_recursive(file->dir);
1491                 remove_subsystem(file->system);
1492                 kmem_cache_free(file_cachep, file);
1493
1494                 /*
1495                  * The do_for_each_event_file_safe() is
1496                  * a double loop. After finding the call for this
1497                  * trace_array, we use break to jump to the next
1498                  * trace_array.
1499                  */
1500                 break;
1501         } while_for_each_event_file();
1502 }
1503
1504 static void event_remove(struct ftrace_event_call *call)
1505 {
1506         struct trace_array *tr;
1507         struct ftrace_event_file *file;
1508
1509         do_for_each_event_file(tr, file) {
1510                 if (file->event_call != call)
1511                         continue;
1512                 ftrace_event_enable_disable(file, 0);
1513                 /*
1514                  * The do_for_each_event_file() is
1515                  * a double loop. After finding the call for this
1516                  * trace_array, we use break to jump to the next
1517                  * trace_array.
1518                  */
1519                 break;
1520         } while_for_each_event_file();
1521
1522         if (call->event.funcs)
1523                 __unregister_ftrace_event(&call->event);
1524         remove_event_from_tracers(call);
1525         list_del(&call->list);
1526 }
1527
1528 static int event_init(struct ftrace_event_call *call)
1529 {
1530         int ret = 0;
1531
1532         if (WARN_ON(!call->name))
1533                 return -EINVAL;
1534
1535         if (call->class->raw_init) {
1536                 ret = call->class->raw_init(call);
1537                 if (ret < 0 && ret != -ENOSYS)
1538                         pr_warn("Could not initialize trace events/%s\n",
1539                                 call->name);
1540         }
1541
1542         return ret;
1543 }
1544
1545 static int
1546 __register_event(struct ftrace_event_call *call, struct module *mod)
1547 {
1548         int ret;
1549
1550         ret = event_init(call);
1551         if (ret < 0)
1552                 return ret;
1553
1554         list_add(&call->list, &ftrace_events);
1555         call->mod = mod;
1556
1557         return 0;
1558 }
1559
1560 static struct ftrace_event_file *
1561 trace_create_new_event(struct ftrace_event_call *call,
1562                        struct trace_array *tr)
1563 {
1564         struct ftrace_event_file *file;
1565
1566         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
1567         if (!file)
1568                 return NULL;
1569
1570         file->event_call = call;
1571         file->tr = tr;
1572         atomic_set(&file->sm_ref, 0);
1573         list_add(&file->list, &tr->events);
1574
1575         return file;
1576 }
1577
1578 /* Add an event to a trace directory */
1579 static int
1580 __trace_add_new_event(struct ftrace_event_call *call,
1581                       struct trace_array *tr,
1582                       const struct file_operations *id,
1583                       const struct file_operations *enable,
1584                       const struct file_operations *filter,
1585                       const struct file_operations *format)
1586 {
1587         struct ftrace_event_file *file;
1588
1589         file = trace_create_new_event(call, tr);
1590         if (!file)
1591                 return -ENOMEM;
1592
1593         return event_create_dir(tr->event_dir, file, id, enable, filter, format);
1594 }
1595
1596 /*
1597  * Just create a decriptor for early init. A descriptor is required
1598  * for enabling events at boot. We want to enable events before
1599  * the filesystem is initialized.
1600  */
1601 static __init int
1602 __trace_early_add_new_event(struct ftrace_event_call *call,
1603                             struct trace_array *tr)
1604 {
1605         struct ftrace_event_file *file;
1606
1607         file = trace_create_new_event(call, tr);
1608         if (!file)
1609                 return -ENOMEM;
1610
1611         return 0;
1612 }
1613
1614 struct ftrace_module_file_ops;
1615 static void __add_event_to_tracers(struct ftrace_event_call *call,
1616                                    struct ftrace_module_file_ops *file_ops);
1617
1618 /* Add an additional event_call dynamically */
1619 int trace_add_event_call(struct ftrace_event_call *call)
1620 {
1621         int ret;
1622         mutex_lock(&trace_types_lock);
1623         mutex_lock(&event_mutex);
1624
1625         ret = __register_event(call, NULL);
1626         if (ret >= 0)
1627                 __add_event_to_tracers(call, NULL);
1628
1629         mutex_unlock(&event_mutex);
1630         mutex_unlock(&trace_types_lock);
1631         return ret;
1632 }
1633
1634 /*
1635  * Must be called under locking of trace_types_lock, event_mutex and
1636  * trace_event_sem.
1637  */
1638 static void __trace_remove_event_call(struct ftrace_event_call *call)
1639 {
1640         event_remove(call);
1641         trace_destroy_fields(call);
1642         destroy_preds(call);
1643 }
1644
1645 /* Remove an event_call */
1646 void trace_remove_event_call(struct ftrace_event_call *call)
1647 {
1648         mutex_lock(&trace_types_lock);
1649         mutex_lock(&event_mutex);
1650         down_write(&trace_event_sem);
1651         __trace_remove_event_call(call);
1652         up_write(&trace_event_sem);
1653         mutex_unlock(&event_mutex);
1654         mutex_unlock(&trace_types_lock);
1655 }
1656
1657 #define for_each_event(event, start, end)                       \
1658         for (event = start;                                     \
1659              (unsigned long)event < (unsigned long)end;         \
1660              event++)
1661
1662 #ifdef CONFIG_MODULES
1663
1664 static LIST_HEAD(ftrace_module_file_list);
1665
1666 /*
1667  * Modules must own their file_operations to keep up with
1668  * reference counting.
1669  */
1670 struct ftrace_module_file_ops {
1671         struct list_head                list;
1672         struct module                   *mod;
1673         struct file_operations          id;
1674         struct file_operations          enable;
1675         struct file_operations          format;
1676         struct file_operations          filter;
1677 };
1678
1679 static struct ftrace_module_file_ops *
1680 find_ftrace_file_ops(struct ftrace_module_file_ops *file_ops, struct module *mod)
1681 {
1682         /*
1683          * As event_calls are added in groups by module,
1684          * when we find one file_ops, we don't need to search for
1685          * each call in that module, as the rest should be the
1686          * same. Only search for a new one if the last one did
1687          * not match.
1688          */
1689         if (file_ops && mod == file_ops->mod)
1690                 return file_ops;
1691
1692         list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1693                 if (file_ops->mod == mod)
1694                         return file_ops;
1695         }
1696         return NULL;
1697 }
1698
1699 static struct ftrace_module_file_ops *
1700 trace_create_file_ops(struct module *mod)
1701 {
1702         struct ftrace_module_file_ops *file_ops;
1703
1704         /*
1705          * This is a bit of a PITA. To allow for correct reference
1706          * counting, modules must "own" their file_operations.
1707          * To do this, we allocate the file operations that will be
1708          * used in the event directory.
1709          */
1710
1711         file_ops = kmalloc(sizeof(*file_ops), GFP_KERNEL);
1712         if (!file_ops)
1713                 return NULL;
1714
1715         file_ops->mod = mod;
1716
1717         file_ops->id = ftrace_event_id_fops;
1718         file_ops->id.owner = mod;
1719
1720         file_ops->enable = ftrace_enable_fops;
1721         file_ops->enable.owner = mod;
1722
1723         file_ops->filter = ftrace_event_filter_fops;
1724         file_ops->filter.owner = mod;
1725
1726         file_ops->format = ftrace_event_format_fops;
1727         file_ops->format.owner = mod;
1728
1729         list_add(&file_ops->list, &ftrace_module_file_list);
1730
1731         return file_ops;
1732 }
1733
1734 static void trace_module_add_events(struct module *mod)
1735 {
1736         struct ftrace_module_file_ops *file_ops = NULL;
1737         struct ftrace_event_call **call, **start, **end;
1738
1739         start = mod->trace_events;
1740         end = mod->trace_events + mod->num_trace_events;
1741
1742         if (start == end)
1743                 return;
1744
1745         file_ops = trace_create_file_ops(mod);
1746         if (!file_ops)
1747                 return;
1748
1749         for_each_event(call, start, end) {
1750                 __register_event(*call, mod);
1751                 __add_event_to_tracers(*call, file_ops);
1752         }
1753 }
1754
1755 static void trace_module_remove_events(struct module *mod)
1756 {
1757         struct ftrace_module_file_ops *file_ops;
1758         struct ftrace_event_call *call, *p;
1759         bool clear_trace = false;
1760
1761         down_write(&trace_event_sem);
1762         list_for_each_entry_safe(call, p, &ftrace_events, list) {
1763                 if (call->mod == mod) {
1764                         if (call->flags & TRACE_EVENT_FL_WAS_ENABLED)
1765                                 clear_trace = true;
1766                         __trace_remove_event_call(call);
1767                 }
1768         }
1769
1770         /* Now free the file_operations */
1771         list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1772                 if (file_ops->mod == mod)
1773                         break;
1774         }
1775         if (&file_ops->list != &ftrace_module_file_list) {
1776                 list_del(&file_ops->list);
1777                 kfree(file_ops);
1778         }
1779         up_write(&trace_event_sem);
1780
1781         /*
1782          * It is safest to reset the ring buffer if the module being unloaded
1783          * registered any events that were used. The only worry is if
1784          * a new module gets loaded, and takes on the same id as the events
1785          * of this module. When printing out the buffer, traced events left
1786          * over from this module may be passed to the new module events and
1787          * unexpected results may occur.
1788          */
1789         if (clear_trace)
1790                 tracing_reset_all_online_cpus();
1791 }
1792
1793 static int trace_module_notify(struct notifier_block *self,
1794                                unsigned long val, void *data)
1795 {
1796         struct module *mod = data;
1797
1798         mutex_lock(&trace_types_lock);
1799         mutex_lock(&event_mutex);
1800         switch (val) {
1801         case MODULE_STATE_COMING:
1802                 trace_module_add_events(mod);
1803                 break;
1804         case MODULE_STATE_GOING:
1805                 trace_module_remove_events(mod);
1806                 break;
1807         }
1808         mutex_unlock(&event_mutex);
1809         mutex_unlock(&trace_types_lock);
1810
1811         return 0;
1812 }
1813
1814 static int
1815 __trace_add_new_mod_event(struct ftrace_event_call *call,
1816                           struct trace_array *tr,
1817                           struct ftrace_module_file_ops *file_ops)
1818 {
1819         return __trace_add_new_event(call, tr,
1820                                      &file_ops->id, &file_ops->enable,
1821                                      &file_ops->filter, &file_ops->format);
1822 }
1823
1824 #else
1825 static inline struct ftrace_module_file_ops *
1826 find_ftrace_file_ops(struct ftrace_module_file_ops *file_ops, struct module *mod)
1827 {
1828         return NULL;
1829 }
1830 static inline int trace_module_notify(struct notifier_block *self,
1831                                       unsigned long val, void *data)
1832 {
1833         return 0;
1834 }
1835 static inline int
1836 __trace_add_new_mod_event(struct ftrace_event_call *call,
1837                           struct trace_array *tr,
1838                           struct ftrace_module_file_ops *file_ops)
1839 {
1840         return -ENODEV;
1841 }
1842 #endif /* CONFIG_MODULES */
1843
1844 /* Create a new event directory structure for a trace directory. */
1845 static void
1846 __trace_add_event_dirs(struct trace_array *tr)
1847 {
1848         struct ftrace_module_file_ops *file_ops = NULL;
1849         struct ftrace_event_call *call;
1850         int ret;
1851
1852         list_for_each_entry(call, &ftrace_events, list) {
1853                 if (call->mod) {
1854                         /*
1855                          * Directories for events by modules need to
1856                          * keep module ref counts when opened (as we don't
1857                          * want the module to disappear when reading one
1858                          * of these files). The file_ops keep account of
1859                          * the module ref count.
1860                          */
1861                         file_ops = find_ftrace_file_ops(file_ops, call->mod);
1862                         if (!file_ops)
1863                                 continue; /* Warn? */
1864                         ret = __trace_add_new_mod_event(call, tr, file_ops);
1865                         if (ret < 0)
1866                                 pr_warning("Could not create directory for event %s\n",
1867                                            call->name);
1868                         continue;
1869                 }
1870                 ret = __trace_add_new_event(call, tr,
1871                                             &ftrace_event_id_fops,
1872                                             &ftrace_enable_fops,
1873                                             &ftrace_event_filter_fops,
1874                                             &ftrace_event_format_fops);
1875                 if (ret < 0)
1876                         pr_warning("Could not create directory for event %s\n",
1877                                    call->name);
1878         }
1879 }
1880
1881 #ifdef CONFIG_DYNAMIC_FTRACE
1882
1883 /* Avoid typos */
1884 #define ENABLE_EVENT_STR        "enable_event"
1885 #define DISABLE_EVENT_STR       "disable_event"
1886
1887 struct event_probe_data {
1888         struct ftrace_event_file        *file;
1889         unsigned long                   count;
1890         int                             ref;
1891         bool                            enable;
1892 };
1893
1894 static struct ftrace_event_file *
1895 find_event_file(struct trace_array *tr, const char *system,  const char *event)
1896 {
1897         struct ftrace_event_file *file;
1898         struct ftrace_event_call *call;
1899
1900         list_for_each_entry(file, &tr->events, list) {
1901
1902                 call = file->event_call;
1903
1904                 if (!call->name || !call->class || !call->class->reg)
1905                         continue;
1906
1907                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
1908                         continue;
1909
1910                 if (strcmp(event, call->name) == 0 &&
1911                     strcmp(system, call->class->system) == 0)
1912                         return file;
1913         }
1914         return NULL;
1915 }
1916
1917 static void
1918 event_enable_probe(unsigned long ip, unsigned long parent_ip, void **_data)
1919 {
1920         struct event_probe_data **pdata = (struct event_probe_data **)_data;
1921         struct event_probe_data *data = *pdata;
1922
1923         if (!data)
1924                 return;
1925
1926         if (data->enable)
1927                 clear_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &data->file->flags);
1928         else
1929                 set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &data->file->flags);
1930 }
1931
1932 static void
1933 event_enable_count_probe(unsigned long ip, unsigned long parent_ip, void **_data)
1934 {
1935         struct event_probe_data **pdata = (struct event_probe_data **)_data;
1936         struct event_probe_data *data = *pdata;
1937
1938         if (!data)
1939                 return;
1940
1941         if (!data->count)
1942                 return;
1943
1944         /* Skip if the event is in a state we want to switch to */
1945         if (data->enable == !(data->file->flags & FTRACE_EVENT_FL_SOFT_DISABLED))
1946                 return;
1947
1948         if (data->count != -1)
1949                 (data->count)--;
1950
1951         event_enable_probe(ip, parent_ip, _data);
1952 }
1953
1954 static int
1955 event_enable_print(struct seq_file *m, unsigned long ip,
1956                       struct ftrace_probe_ops *ops, void *_data)
1957 {
1958         struct event_probe_data *data = _data;
1959
1960         seq_printf(m, "%ps:", (void *)ip);
1961
1962         seq_printf(m, "%s:%s:%s",
1963                    data->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
1964                    data->file->event_call->class->system,
1965                    data->file->event_call->name);
1966
1967         if (data->count == -1)
1968                 seq_printf(m, ":unlimited\n");
1969         else
1970                 seq_printf(m, ":count=%ld\n", data->count);
1971
1972         return 0;
1973 }
1974
1975 static int
1976 event_enable_init(struct ftrace_probe_ops *ops, unsigned long ip,
1977                   void **_data)
1978 {
1979         struct event_probe_data **pdata = (struct event_probe_data **)_data;
1980         struct event_probe_data *data = *pdata;
1981
1982         data->ref++;
1983         return 0;
1984 }
1985
1986 static void
1987 event_enable_free(struct ftrace_probe_ops *ops, unsigned long ip,
1988                   void **_data)
1989 {
1990         struct event_probe_data **pdata = (struct event_probe_data **)_data;
1991         struct event_probe_data *data = *pdata;
1992
1993         if (WARN_ON_ONCE(data->ref <= 0))
1994                 return;
1995
1996         data->ref--;
1997         if (!data->ref) {
1998                 /* Remove the SOFT_MODE flag */
1999                 __ftrace_event_enable_disable(data->file, 0, 1);
2000                 module_put(data->file->event_call->mod);
2001                 kfree(data);
2002         }
2003         *pdata = NULL;
2004 }
2005
2006 static struct ftrace_probe_ops event_enable_probe_ops = {
2007         .func                   = event_enable_probe,
2008         .print                  = event_enable_print,
2009         .init                   = event_enable_init,
2010         .free                   = event_enable_free,
2011 };
2012
2013 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2014         .func                   = event_enable_count_probe,
2015         .print                  = event_enable_print,
2016         .init                   = event_enable_init,
2017         .free                   = event_enable_free,
2018 };
2019
2020 static struct ftrace_probe_ops event_disable_probe_ops = {
2021         .func                   = event_enable_probe,
2022         .print                  = event_enable_print,
2023         .init                   = event_enable_init,
2024         .free                   = event_enable_free,
2025 };
2026
2027 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2028         .func                   = event_enable_count_probe,
2029         .print                  = event_enable_print,
2030         .init                   = event_enable_init,
2031         .free                   = event_enable_free,
2032 };
2033
2034 static int
2035 event_enable_func(struct ftrace_hash *hash,
2036                   char *glob, char *cmd, char *param, int enabled)
2037 {
2038         struct trace_array *tr = top_trace_array();
2039         struct ftrace_event_file *file;
2040         struct ftrace_probe_ops *ops;
2041         struct event_probe_data *data;
2042         const char *system;
2043         const char *event;
2044         char *number;
2045         bool enable;
2046         int ret;
2047
2048         /* hash funcs only work with set_ftrace_filter */
2049         if (!enabled || !param)
2050                 return -EINVAL;
2051
2052         system = strsep(&param, ":");
2053         if (!param)
2054                 return -EINVAL;
2055
2056         event = strsep(&param, ":");
2057
2058         mutex_lock(&event_mutex);
2059
2060         ret = -EINVAL;
2061         file = find_event_file(tr, system, event);
2062         if (!file)
2063                 goto out;
2064
2065         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2066
2067         if (enable)
2068                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2069         else
2070                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2071
2072         if (glob[0] == '!') {
2073                 unregister_ftrace_function_probe_func(glob+1, ops);
2074                 ret = 0;
2075                 goto out;
2076         }
2077
2078         ret = -ENOMEM;
2079         data = kzalloc(sizeof(*data), GFP_KERNEL);
2080         if (!data)
2081                 goto out;
2082
2083         data->enable = enable;
2084         data->count = -1;
2085         data->file = file;
2086
2087         if (!param)
2088                 goto out_reg;
2089
2090         number = strsep(&param, ":");
2091
2092         ret = -EINVAL;
2093         if (!strlen(number))
2094                 goto out_free;
2095
2096         /*
2097          * We use the callback data field (which is a pointer)
2098          * as our counter.
2099          */
2100         ret = kstrtoul(number, 0, &data->count);
2101         if (ret)
2102                 goto out_free;
2103
2104  out_reg:
2105         /* Don't let event modules unload while probe registered */
2106         ret = try_module_get(file->event_call->mod);
2107         if (!ret) {
2108                 ret = -EBUSY;
2109                 goto out_free;
2110         }
2111
2112         ret = __ftrace_event_enable_disable(file, 1, 1);
2113         if (ret < 0)
2114                 goto out_put;
2115         ret = register_ftrace_function_probe(glob, ops, data);
2116         /*
2117          * The above returns on success the # of functions enabled,
2118          * but if it didn't find any functions it returns zero.
2119          * Consider no functions a failure too.
2120          */
2121         if (!ret) {
2122                 ret = -ENOENT;
2123                 goto out_disable;
2124         } else if (ret < 0)
2125                 goto out_disable;
2126         /* Just return zero, not the number of enabled functions */
2127         ret = 0;
2128  out:
2129         mutex_unlock(&event_mutex);
2130         return ret;
2131
2132  out_disable:
2133         __ftrace_event_enable_disable(file, 0, 1);
2134  out_put:
2135         module_put(file->event_call->mod);
2136  out_free:
2137         kfree(data);
2138         goto out;
2139 }
2140
2141 static struct ftrace_func_command event_enable_cmd = {
2142         .name                   = ENABLE_EVENT_STR,
2143         .func                   = event_enable_func,
2144 };
2145
2146 static struct ftrace_func_command event_disable_cmd = {
2147         .name                   = DISABLE_EVENT_STR,
2148         .func                   = event_enable_func,
2149 };
2150
2151 static __init int register_event_cmds(void)
2152 {
2153         int ret;
2154
2155         ret = register_ftrace_command(&event_enable_cmd);
2156         if (WARN_ON(ret < 0))
2157                 return ret;
2158         ret = register_ftrace_command(&event_disable_cmd);
2159         if (WARN_ON(ret < 0))
2160                 unregister_ftrace_command(&event_enable_cmd);
2161         return ret;
2162 }
2163 #else
2164 static inline int register_event_cmds(void) { return 0; }
2165 #endif /* CONFIG_DYNAMIC_FTRACE */
2166
2167 /*
2168  * The top level array has already had its ftrace_event_file
2169  * descriptors created in order to allow for early events to
2170  * be recorded. This function is called after the debugfs has been
2171  * initialized, and we now have to create the files associated
2172  * to the events.
2173  */
2174 static __init void
2175 __trace_early_add_event_dirs(struct trace_array *tr)
2176 {
2177         struct ftrace_event_file *file;
2178         int ret;
2179
2180
2181         list_for_each_entry(file, &tr->events, list) {
2182                 ret = event_create_dir(tr->event_dir, file,
2183                                        &ftrace_event_id_fops,
2184                                        &ftrace_enable_fops,
2185                                        &ftrace_event_filter_fops,
2186                                        &ftrace_event_format_fops);
2187                 if (ret < 0)
2188                         pr_warning("Could not create directory for event %s\n",
2189                                    file->event_call->name);
2190         }
2191 }
2192
2193 /*
2194  * For early boot up, the top trace array requires to have
2195  * a list of events that can be enabled. This must be done before
2196  * the filesystem is set up in order to allow events to be traced
2197  * early.
2198  */
2199 static __init void
2200 __trace_early_add_events(struct trace_array *tr)
2201 {
2202         struct ftrace_event_call *call;
2203         int ret;
2204
2205         list_for_each_entry(call, &ftrace_events, list) {
2206                 /* Early boot up should not have any modules loaded */
2207                 if (WARN_ON_ONCE(call->mod))
2208                         continue;
2209
2210                 ret = __trace_early_add_new_event(call, tr);
2211                 if (ret < 0)
2212                         pr_warning("Could not create early event %s\n",
2213                                    call->name);
2214         }
2215 }
2216
2217 /* Remove the event directory structure for a trace directory. */
2218 static void
2219 __trace_remove_event_dirs(struct trace_array *tr)
2220 {
2221         struct ftrace_event_file *file, *next;
2222
2223         list_for_each_entry_safe(file, next, &tr->events, list) {
2224                 list_del(&file->list);
2225                 debugfs_remove_recursive(file->dir);
2226                 remove_subsystem(file->system);
2227                 kmem_cache_free(file_cachep, file);
2228         }
2229 }
2230
2231 static void
2232 __add_event_to_tracers(struct ftrace_event_call *call,
2233                        struct ftrace_module_file_ops *file_ops)
2234 {
2235         struct trace_array *tr;
2236
2237         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2238                 if (file_ops)
2239                         __trace_add_new_mod_event(call, tr, file_ops);
2240                 else
2241                         __trace_add_new_event(call, tr,
2242                                               &ftrace_event_id_fops,
2243                                               &ftrace_enable_fops,
2244                                               &ftrace_event_filter_fops,
2245                                               &ftrace_event_format_fops);
2246         }
2247 }
2248
2249 static struct notifier_block trace_module_nb = {
2250         .notifier_call = trace_module_notify,
2251         .priority = 0,
2252 };
2253
2254 extern struct ftrace_event_call *__start_ftrace_events[];
2255 extern struct ftrace_event_call *__stop_ftrace_events[];
2256
2257 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2258
2259 static __init int setup_trace_event(char *str)
2260 {
2261         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2262         ring_buffer_expanded = true;
2263         tracing_selftest_disabled = true;
2264
2265         return 1;
2266 }
2267 __setup("trace_event=", setup_trace_event);
2268
2269 /* Expects to have event_mutex held when called */
2270 static int
2271 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2272 {
2273         struct dentry *d_events;
2274         struct dentry *entry;
2275
2276         entry = debugfs_create_file("set_event", 0644, parent,
2277                                     tr, &ftrace_set_event_fops);
2278         if (!entry) {
2279                 pr_warning("Could not create debugfs 'set_event' entry\n");
2280                 return -ENOMEM;
2281         }
2282
2283         d_events = debugfs_create_dir("events", parent);
2284         if (!d_events) {
2285                 pr_warning("Could not create debugfs 'events' directory\n");
2286                 return -ENOMEM;
2287         }
2288
2289         /* ring buffer internal formats */
2290         trace_create_file("header_page", 0444, d_events,
2291                           ring_buffer_print_page_header,
2292                           &ftrace_show_header_fops);
2293
2294         trace_create_file("header_event", 0444, d_events,
2295                           ring_buffer_print_entry_header,
2296                           &ftrace_show_header_fops);
2297
2298         trace_create_file("enable", 0644, d_events,
2299                           tr, &ftrace_tr_enable_fops);
2300
2301         tr->event_dir = d_events;
2302
2303         return 0;
2304 }
2305
2306 /**
2307  * event_trace_add_tracer - add a instance of a trace_array to events
2308  * @parent: The parent dentry to place the files/directories for events in
2309  * @tr: The trace array associated with these events
2310  *
2311  * When a new instance is created, it needs to set up its events
2312  * directory, as well as other files associated with events. It also
2313  * creates the event hierachry in the @parent/events directory.
2314  *
2315  * Returns 0 on success.
2316  */
2317 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2318 {
2319         int ret;
2320
2321         mutex_lock(&event_mutex);
2322
2323         ret = create_event_toplevel_files(parent, tr);
2324         if (ret)
2325                 goto out_unlock;
2326
2327         down_write(&trace_event_sem);
2328         __trace_add_event_dirs(tr);
2329         up_write(&trace_event_sem);
2330
2331  out_unlock:
2332         mutex_unlock(&event_mutex);
2333
2334         return ret;
2335 }
2336
2337 /*
2338  * The top trace array already had its file descriptors created.
2339  * Now the files themselves need to be created.
2340  */
2341 static __init int
2342 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
2343 {
2344         int ret;
2345
2346         mutex_lock(&event_mutex);
2347
2348         ret = create_event_toplevel_files(parent, tr);
2349         if (ret)
2350                 goto out_unlock;
2351
2352         down_write(&trace_event_sem);
2353         __trace_early_add_event_dirs(tr);
2354         up_write(&trace_event_sem);
2355
2356  out_unlock:
2357         mutex_unlock(&event_mutex);
2358
2359         return ret;
2360 }
2361
2362 int event_trace_del_tracer(struct trace_array *tr)
2363 {
2364         /* Disable any running events */
2365         __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
2366
2367         mutex_lock(&event_mutex);
2368
2369         down_write(&trace_event_sem);
2370         __trace_remove_event_dirs(tr);
2371         debugfs_remove_recursive(tr->event_dir);
2372         up_write(&trace_event_sem);
2373
2374         tr->event_dir = NULL;
2375
2376         mutex_unlock(&event_mutex);
2377
2378         return 0;
2379 }
2380
2381 static __init int event_trace_memsetup(void)
2382 {
2383         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
2384         file_cachep = KMEM_CACHE(ftrace_event_file, SLAB_PANIC);
2385         return 0;
2386 }
2387
2388 static __init int event_trace_enable(void)
2389 {
2390         struct trace_array *tr = top_trace_array();
2391         struct ftrace_event_call **iter, *call;
2392         char *buf = bootup_event_buf;
2393         char *token;
2394         int ret;
2395
2396         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
2397
2398                 call = *iter;
2399                 ret = event_init(call);
2400                 if (!ret)
2401                         list_add(&call->list, &ftrace_events);
2402         }
2403
2404         /*
2405          * We need the top trace array to have a working set of trace
2406          * points at early init, before the debug files and directories
2407          * are created. Create the file entries now, and attach them
2408          * to the actual file dentries later.
2409          */
2410         __trace_early_add_events(tr);
2411
2412         while (true) {
2413                 token = strsep(&buf, ",");
2414
2415                 if (!token)
2416                         break;
2417                 if (!*token)
2418                         continue;
2419
2420                 ret = ftrace_set_clr_event(tr, token, 1);
2421                 if (ret)
2422                         pr_warn("Failed to enable trace event: %s\n", token);
2423         }
2424
2425         trace_printk_start_comm();
2426
2427         register_event_cmds();
2428
2429         return 0;
2430 }
2431
2432 static __init int event_trace_init(void)
2433 {
2434         struct trace_array *tr;
2435         struct dentry *d_tracer;
2436         struct dentry *entry;
2437         int ret;
2438
2439         tr = top_trace_array();
2440
2441         d_tracer = tracing_init_dentry();
2442         if (!d_tracer)
2443                 return 0;
2444
2445         entry = debugfs_create_file("available_events", 0444, d_tracer,
2446                                     tr, &ftrace_avail_fops);
2447         if (!entry)
2448                 pr_warning("Could not create debugfs "
2449                            "'available_events' entry\n");
2450
2451         if (trace_define_common_fields())
2452                 pr_warning("tracing: Failed to allocate common fields");
2453
2454         ret = early_event_add_tracer(d_tracer, tr);
2455         if (ret)
2456                 return ret;
2457
2458         ret = register_module_notifier(&trace_module_nb);
2459         if (ret)
2460                 pr_warning("Failed to register trace events module notifier\n");
2461
2462         return 0;
2463 }
2464 early_initcall(event_trace_memsetup);
2465 core_initcall(event_trace_enable);
2466 fs_initcall(event_trace_init);
2467
2468 #ifdef CONFIG_FTRACE_STARTUP_TEST
2469
2470 static DEFINE_SPINLOCK(test_spinlock);
2471 static DEFINE_SPINLOCK(test_spinlock_irq);
2472 static DEFINE_MUTEX(test_mutex);
2473
2474 static __init void test_work(struct work_struct *dummy)
2475 {
2476         spin_lock(&test_spinlock);
2477         spin_lock_irq(&test_spinlock_irq);
2478         udelay(1);
2479         spin_unlock_irq(&test_spinlock_irq);
2480         spin_unlock(&test_spinlock);
2481
2482         mutex_lock(&test_mutex);
2483         msleep(1);
2484         mutex_unlock(&test_mutex);
2485 }
2486
2487 static __init int event_test_thread(void *unused)
2488 {
2489         void *test_malloc;
2490
2491         test_malloc = kmalloc(1234, GFP_KERNEL);
2492         if (!test_malloc)
2493                 pr_info("failed to kmalloc\n");
2494
2495         schedule_on_each_cpu(test_work);
2496
2497         kfree(test_malloc);
2498
2499         set_current_state(TASK_INTERRUPTIBLE);
2500         while (!kthread_should_stop())
2501                 schedule();
2502
2503         return 0;
2504 }
2505
2506 /*
2507  * Do various things that may trigger events.
2508  */
2509 static __init void event_test_stuff(void)
2510 {
2511         struct task_struct *test_thread;
2512
2513         test_thread = kthread_run(event_test_thread, NULL, "test-events");
2514         msleep(1);
2515         kthread_stop(test_thread);
2516 }
2517
2518 /*
2519  * For every trace event defined, we will test each trace point separately,
2520  * and then by groups, and finally all trace points.
2521  */
2522 static __init void event_trace_self_tests(void)
2523 {
2524         struct ftrace_subsystem_dir *dir;
2525         struct ftrace_event_file *file;
2526         struct ftrace_event_call *call;
2527         struct event_subsystem *system;
2528         struct trace_array *tr;
2529         int ret;
2530
2531         tr = top_trace_array();
2532
2533         pr_info("Running tests on trace events:\n");
2534
2535         list_for_each_entry(file, &tr->events, list) {
2536
2537                 call = file->event_call;
2538
2539                 /* Only test those that have a probe */
2540                 if (!call->class || !call->class->probe)
2541                         continue;
2542
2543 /*
2544  * Testing syscall events here is pretty useless, but
2545  * we still do it if configured. But this is time consuming.
2546  * What we really need is a user thread to perform the
2547  * syscalls as we test.
2548  */
2549 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
2550                 if (call->class->system &&
2551                     strcmp(call->class->system, "syscalls") == 0)
2552                         continue;
2553 #endif
2554
2555                 pr_info("Testing event %s: ", call->name);
2556
2557                 /*
2558                  * If an event is already enabled, someone is using
2559                  * it and the self test should not be on.
2560                  */
2561                 if (file->flags & FTRACE_EVENT_FL_ENABLED) {
2562                         pr_warning("Enabled event during self test!\n");
2563                         WARN_ON_ONCE(1);
2564                         continue;
2565                 }
2566
2567                 ftrace_event_enable_disable(file, 1);
2568                 event_test_stuff();
2569                 ftrace_event_enable_disable(file, 0);
2570
2571                 pr_cont("OK\n");
2572         }
2573
2574         /* Now test at the sub system level */
2575
2576         pr_info("Running tests on trace event systems:\n");
2577
2578         list_for_each_entry(dir, &tr->systems, list) {
2579
2580                 system = dir->subsystem;
2581
2582                 /* the ftrace system is special, skip it */
2583                 if (strcmp(system->name, "ftrace") == 0)
2584                         continue;
2585
2586                 pr_info("Testing event system %s: ", system->name);
2587
2588                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
2589                 if (WARN_ON_ONCE(ret)) {
2590                         pr_warning("error enabling system %s\n",
2591                                    system->name);
2592                         continue;
2593                 }
2594
2595                 event_test_stuff();
2596
2597                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
2598                 if (WARN_ON_ONCE(ret)) {
2599                         pr_warning("error disabling system %s\n",
2600                                    system->name);
2601                         continue;
2602                 }
2603
2604                 pr_cont("OK\n");
2605         }
2606
2607         /* Test with all events enabled */
2608
2609         pr_info("Running tests on all trace events:\n");
2610         pr_info("Testing all events: ");
2611
2612         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
2613         if (WARN_ON_ONCE(ret)) {
2614                 pr_warning("error enabling all events\n");
2615                 return;
2616         }
2617
2618         event_test_stuff();
2619
2620         /* reset sysname */
2621         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
2622         if (WARN_ON_ONCE(ret)) {
2623                 pr_warning("error disabling all events\n");
2624                 return;
2625         }
2626
2627         pr_cont("OK\n");
2628 }
2629
2630 #ifdef CONFIG_FUNCTION_TRACER
2631
2632 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
2633
2634 static void
2635 function_test_events_call(unsigned long ip, unsigned long parent_ip,
2636                           struct ftrace_ops *op, struct pt_regs *pt_regs)
2637 {
2638         struct ring_buffer_event *event;
2639         struct ring_buffer *buffer;
2640         struct ftrace_entry *entry;
2641         unsigned long flags;
2642         long disabled;
2643         int cpu;
2644         int pc;
2645
2646         pc = preempt_count();
2647         preempt_disable_notrace();
2648         cpu = raw_smp_processor_id();
2649         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
2650
2651         if (disabled != 1)
2652                 goto out;
2653
2654         local_save_flags(flags);
2655
2656         event = trace_current_buffer_lock_reserve(&buffer,
2657                                                   TRACE_FN, sizeof(*entry),
2658                                                   flags, pc);
2659         if (!event)
2660                 goto out;
2661         entry   = ring_buffer_event_data(event);
2662         entry->ip                       = ip;
2663         entry->parent_ip                = parent_ip;
2664
2665         trace_buffer_unlock_commit(buffer, event, flags, pc);
2666
2667  out:
2668         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
2669         preempt_enable_notrace();
2670 }
2671
2672 static struct ftrace_ops trace_ops __initdata  =
2673 {
2674         .func = function_test_events_call,
2675         .flags = FTRACE_OPS_FL_RECURSION_SAFE,
2676 };
2677
2678 static __init void event_trace_self_test_with_function(void)
2679 {
2680         int ret;
2681         ret = register_ftrace_function(&trace_ops);
2682         if (WARN_ON(ret < 0)) {
2683                 pr_info("Failed to enable function tracer for event tests\n");
2684                 return;
2685         }
2686         pr_info("Running tests again, along with the function tracer\n");
2687         event_trace_self_tests();
2688         unregister_ftrace_function(&trace_ops);
2689 }
2690 #else
2691 static __init void event_trace_self_test_with_function(void)
2692 {
2693 }
2694 #endif
2695
2696 static __init int event_trace_self_tests_init(void)
2697 {
2698         if (!tracing_selftest_disabled) {
2699                 event_trace_self_tests();
2700                 event_trace_self_test_with_function();
2701         }
2702
2703         return 0;
2704 }
2705
2706 late_initcall(event_trace_self_tests_init);
2707
2708 #endif