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