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 ftrace_event_field *field;
 830         struct list_head *common_head = &ftrace_common_fields;
 831         struct list_head *head = trace_get_fields(call);
 832
 833         (*pos)++;
 834
 835         switch ((unsigned long)v) {
 836         case FORMAT_HEADER:
 837                 if (unlikely(list_empty(common_head)))
 838                         return NULL;
 839
 840                 field = list_entry(common_head->prev,
 841                                    struct ftrace_event_field, link);
 842                 return field;
 843
 844         case FORMAT_FIELD_SEPERATOR:
 845                 if (unlikely(list_empty(head)))
 846                         return NULL;
 847
 848                 field = list_entry(head->prev, struct ftrace_event_field, link);
 849                 return field;
 850
 851         case FORMAT_PRINTFMT:
 852                 /* all done */
 853                 return NULL;
 854         }
 855
 856         field = v;
 857         if (field->link.prev == common_head)
 858                 return (void *)FORMAT_FIELD_SEPERATOR;
 859         else if (field->link.prev == head)
 860                 return (void *)FORMAT_PRINTFMT;
 861
 862         field = list_entry(field->link.prev, struct ftrace_event_field, link);
 863
 864         return field;
 865 }
 866
 867 static void *f_start(struct seq_file *m, loff_t *pos)
 868 {
 869         loff_t l = 0;
 870         void *p;
 871
 872         /* Start by showing the header */
 873         if (!*pos)
 874                 return (void *)FORMAT_HEADER;
 875
 876         p = (void *)FORMAT_HEADER;
 877         do {
 878                 p = f_next(m, p, &l);
 879         } while (p && l < *pos);
 880
 881         return p;
 882 }
 883
 884 static int f_show(struct seq_file *m, void *v)
 885 {
 886         struct ftrace_event_call *call = m->private;
 887         struct ftrace_event_field *field;
 888         const char *array_descriptor;
 889
 890         switch ((unsigned long)v) {
 891         case FORMAT_HEADER:
 892                 seq_printf(m, "name: %s\n", call->name);
 893                 seq_printf(m, "ID: %d\n", call->event.type);
 894                 seq_printf(m, "format:\n");
 895                 return 0;
 896
 897         case FORMAT_FIELD_SEPERATOR:
 898                 seq_putc(m, '\n');
 899                 return 0;
 900
 901         case FORMAT_PRINTFMT:
 902                 seq_printf(m, "\nprint fmt: %s\n",
 903                            call->print_fmt);
 904                 return 0;
 905         }
 906
 907         field = v;
 908
 909         /*
 910          * Smartly shows the array type(except dynamic array).
 911          * Normal:
 912          *      field:TYPE VAR
 913          * If TYPE := TYPE[LEN], it is shown:
 914          *      field:TYPE VAR[LEN]
 915          */
 916         array_descriptor = strchr(field->type, '[');
 917
 918         if (!strncmp(field->type, "__data_loc", 10))
 919                 array_descriptor = NULL;
 920
 921         if (!array_descriptor)
 922                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
 923                            field->type, field->name, field->offset,
 924                            field->size, !!field->is_signed);
 925         else
 926                 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
 927                            (int)(array_descriptor - field->type),
 928                            field->type, field->name,
 929                            array_descriptor, field->offset,
 930                            field->size, !!field->is_signed);
 931
 932         return 0;
 933 }
 934
 935 static void f_stop(struct seq_file *m, void *p)
 936 {
 937 }
 938
 939 static const struct seq_operations trace_format_seq_ops = {
 940         .start          = f_start,
 941         .next           = f_next,
 942         .stop           = f_stop,
 943         .show           = f_show,
 944 };
 945
 946 static int trace_format_open(struct inode *inode, struct file *file)
 947 {
 948         struct ftrace_event_call *call = inode->i_private;
 949         struct seq_file *m;
 950         int ret;
 951
 952         ret = seq_open(file, &trace_format_seq_ops);
 953         if (ret < 0)
 954                 return ret;
 955
 956         m = file->private_data;
 957         m->private = call;
 958
 959         return 0;
 960 }
 961
 962 static ssize_t
 963 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 964 {
 965         struct ftrace_event_call *call = filp->private_data;
 966         struct trace_seq *s;
 967         int r;
 968
 969         if (*ppos)
 970                 return 0;
 971
 972         s = kmalloc(sizeof(*s), GFP_KERNEL);
 973         if (!s)
 974                 return -ENOMEM;
 975
 976         trace_seq_init(s);
 977         trace_seq_printf(s, "%d\n", call->event.type);
 978
 979         r = simple_read_from_buffer(ubuf, cnt, ppos,
 980                                     s->buffer, s->len);
 981         kfree(s);
 982         return r;
 983 }
 984
 985 static ssize_t
 986 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 987                   loff_t *ppos)
 988 {
 989         struct ftrace_event_call *call = filp->private_data;
 990         struct trace_seq *s;
 991         int r;
 992
 993         if (*ppos)
 994                 return 0;
 995
 996         s = kmalloc(sizeof(*s), GFP_KERNEL);
 997         if (!s)
 998                 return -ENOMEM;
 999
1000         trace_seq_init(s);
1001
1002         print_event_filter(call, s);
1003         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
1004
1005         kfree(s);
1006
1007         return r;
1008 }
1009
1010 static ssize_t
1011 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1012                    loff_t *ppos)
1013 {
1014         struct ftrace_event_call *call = filp->private_data;
1015         char *buf;
1016         int err;
1017
1018         if (cnt >= PAGE_SIZE)
1019                 return -EINVAL;
1020
1021         buf = (char *)__get_free_page(GFP_TEMPORARY);
1022         if (!buf)
1023                 return -ENOMEM;
1024
1025         if (copy_from_user(buf, ubuf, cnt)) {
1026                 free_page((unsigned long) buf);
1027                 return -EFAULT;
1028         }
1029         buf[cnt] = '\0';
1030
1031         err = apply_event_filter(call, buf);
1032         free_page((unsigned long) buf);
1033         if (err < 0)
1034                 return err;
1035
1036         *ppos += cnt;
1037
1038         return cnt;
1039 }
1040
1041 static LIST_HEAD(event_subsystems);
1042
1043 static int subsystem_open(struct inode *inode, struct file *filp)
1044 {
1045         struct event_subsystem *system = NULL;
1046         struct ftrace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1047         struct trace_array *tr;
1048         int ret;
1049
1050         /* Make sure the system still exists */
1051         mutex_lock(&trace_types_lock);
1052         mutex_lock(&event_mutex);
1053         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1054                 list_for_each_entry(dir, &tr->systems, list) {
1055                         if (dir == inode->i_private) {
1056                                 /* Don't open systems with no events */
1057                                 if (dir->nr_events) {
1058                                         __get_system_dir(dir);
1059                                         system = dir->subsystem;
1060                                 }
1061                                 goto exit_loop;
1062                         }
1063                 }
1064         }
1065  exit_loop:
1066         mutex_unlock(&event_mutex);
1067         mutex_unlock(&trace_types_lock);
1068
1069         if (!system)
1070                 return -ENODEV;
1071
1072         /* Some versions of gcc think dir can be uninitialized here */
1073         WARN_ON(!dir);
1074
1075         /* Still need to increment the ref count of the system */
1076         if (trace_array_get(tr) < 0) {
1077                 put_system(dir);
1078                 return -ENODEV;
1079         }
1080
1081         ret = tracing_open_generic(inode, filp);
1082         if (ret < 0) {
1083                 trace_array_put(tr);
1084                 put_system(dir);
1085         }
1086
1087         return ret;
1088 }
1089
1090 static int system_tr_open(struct inode *inode, struct file *filp)
1091 {
1092         struct ftrace_subsystem_dir *dir;
1093         struct trace_array *tr = inode->i_private;
1094         int ret;
1095
1096         if (trace_array_get(tr) < 0)
1097                 return -ENODEV;
1098
1099         /* Make a temporary dir that has no system but points to tr */
1100         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1101         if (!dir) {
1102                 trace_array_put(tr);
1103                 return -ENOMEM;
1104         }
1105
1106         dir->tr = tr;
1107
1108         ret = tracing_open_generic(inode, filp);
1109         if (ret < 0) {
1110                 trace_array_put(tr);
1111                 kfree(dir);
1112         }
1113
1114         filp->private_data = dir;
1115
1116         return ret;
1117 }
1118
1119 static int subsystem_release(struct inode *inode, struct file *file)
1120 {
1121         struct ftrace_subsystem_dir *dir = file->private_data;
1122
1123         trace_array_put(dir->tr);
1124
1125         /*
1126          * If dir->subsystem is NULL, then this is a temporary
1127          * descriptor that was made for a trace_array to enable
1128          * all subsystems.
1129          */
1130         if (dir->subsystem)
1131                 put_system(dir);
1132         else
1133                 kfree(dir);
1134
1135         return 0;
1136 }
1137
1138 static ssize_t
1139 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1140                       loff_t *ppos)
1141 {
1142         struct ftrace_subsystem_dir *dir = filp->private_data;
1143         struct event_subsystem *system = dir->subsystem;
1144         struct trace_seq *s;
1145         int r;
1146
1147         if (*ppos)
1148                 return 0;
1149
1150         s = kmalloc(sizeof(*s), GFP_KERNEL);
1151         if (!s)
1152                 return -ENOMEM;
1153
1154         trace_seq_init(s);
1155
1156         print_subsystem_event_filter(system, s);
1157         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
1158
1159         kfree(s);
1160
1161         return r;
1162 }
1163
1164 static ssize_t
1165 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1166                        loff_t *ppos)
1167 {
1168         struct ftrace_subsystem_dir *dir = filp->private_data;
1169         char *buf;
1170         int err;
1171
1172         if (cnt >= PAGE_SIZE)
1173                 return -EINVAL;
1174
1175         buf = (char *)__get_free_page(GFP_TEMPORARY);
1176         if (!buf)
1177                 return -ENOMEM;
1178
1179         if (copy_from_user(buf, ubuf, cnt)) {
1180                 free_page((unsigned long) buf);
1181                 return -EFAULT;
1182         }
1183         buf[cnt] = '\0';
1184
1185         err = apply_subsystem_event_filter(dir, buf);
1186         free_page((unsigned long) buf);
1187         if (err < 0)
1188                 return err;
1189
1190         *ppos += cnt;
1191
1192         return cnt;
1193 }
1194
1195 static ssize_t
1196 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1197 {
1198         int (*func)(struct trace_seq *s) = filp->private_data;
1199         struct trace_seq *s;
1200         int r;
1201
1202         if (*ppos)
1203                 return 0;
1204
1205         s = kmalloc(sizeof(*s), GFP_KERNEL);
1206         if (!s)
1207                 return -ENOMEM;
1208
1209         trace_seq_init(s);
1210
1211         func(s);
1212         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
1213
1214         kfree(s);
1215
1216         return r;
1217 }
1218
1219 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1220 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1221
1222 static const struct seq_operations show_event_seq_ops = {
1223         .start = t_start,
1224         .next = t_next,
1225         .show = t_show,
1226         .stop = t_stop,
1227 };
1228
1229 static const struct seq_operations show_set_event_seq_ops = {
1230         .start = s_start,
1231         .next = s_next,
1232         .show = t_show,
1233         .stop = t_stop,
1234 };
1235
1236 static const struct file_operations ftrace_avail_fops = {
1237         .open = ftrace_event_avail_open,
1238         .read = seq_read,
1239         .llseek = seq_lseek,
1240         .release = seq_release,
1241 };
1242
1243 static const struct file_operations ftrace_set_event_fops = {
1244         .open = ftrace_event_set_open,
1245         .read = seq_read,
1246         .write = ftrace_event_write,
1247         .llseek = seq_lseek,
1248         .release = seq_release,
1249 };
1250
1251 static const struct file_operations ftrace_enable_fops = {
1252         .open = tracing_open_generic_file,
1253         .read = event_enable_read,
1254         .write = event_enable_write,
1255         .release = tracing_release_generic_file,
1256         .llseek = default_llseek,
1257 };
1258
1259 static const struct file_operations ftrace_event_format_fops = {
1260         .open = trace_format_open,
1261         .read = seq_read,
1262         .llseek = seq_lseek,
1263         .release = seq_release,
1264 };
1265
1266 static const struct file_operations ftrace_event_id_fops = {
1267         .open = tracing_open_generic,
1268         .read = event_id_read,
1269         .llseek = default_llseek,
1270 };
1271
1272 static const struct file_operations ftrace_event_filter_fops = {
1273         .open = tracing_open_generic,
1274         .read = event_filter_read,
1275         .write = event_filter_write,
1276         .llseek = default_llseek,
1277 };
1278
1279 static const struct file_operations ftrace_subsystem_filter_fops = {
1280         .open = subsystem_open,
1281         .read = subsystem_filter_read,
1282         .write = subsystem_filter_write,
1283         .llseek = default_llseek,
1284         .release = subsystem_release,
1285 };
1286
1287 static const struct file_operations ftrace_system_enable_fops = {
1288         .open = subsystem_open,
1289         .read = system_enable_read,
1290         .write = system_enable_write,
1291         .llseek = default_llseek,
1292         .release = subsystem_release,
1293 };
1294
1295 static const struct file_operations ftrace_tr_enable_fops = {
1296         .open = system_tr_open,
1297         .read = system_enable_read,
1298         .write = system_enable_write,
1299         .llseek = default_llseek,
1300         .release = subsystem_release,
1301 };
1302
1303 static const struct file_operations ftrace_show_header_fops = {
1304         .open = tracing_open_generic,
1305         .read = show_header,
1306         .llseek = default_llseek,
1307 };
1308
1309 static int
1310 ftrace_event_open(struct inode *inode, struct file *file,
1311                   const struct seq_operations *seq_ops)
1312 {
1313         struct seq_file *m;
1314         int ret;
1315
1316         ret = seq_open(file, seq_ops);
1317         if (ret < 0)
1318                 return ret;
1319         m = file->private_data;
1320         /* copy tr over to seq ops */
1321         m->private = inode->i_private;
1322
1323         return ret;
1324 }
1325
1326 static int
1327 ftrace_event_avail_open(struct inode *inode, struct file *file)
1328 {
1329         const struct seq_operations *seq_ops = &show_event_seq_ops;
1330
1331         return ftrace_event_open(inode, file, seq_ops);
1332 }
1333
1334 static int
1335 ftrace_event_set_open(struct inode *inode, struct file *file)
1336 {
1337         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1338         struct trace_array *tr = inode->i_private;
1339
1340         if ((file->f_mode & FMODE_WRITE) &&
1341             (file->f_flags & O_TRUNC))
1342                 ftrace_clear_events(tr);
1343
1344         return ftrace_event_open(inode, file, seq_ops);
1345 }
1346
1347 static struct event_subsystem *
1348 create_new_subsystem(const char *name)
1349 {
1350         struct event_subsystem *system;
1351
1352         /* need to create new entry */
1353         system = kmalloc(sizeof(*system), GFP_KERNEL);
1354         if (!system)
1355                 return NULL;
1356
1357         system->ref_count = 1;
1358
1359         /* Only allocate if dynamic (kprobes and modules) */
1360         if (!core_kernel_data((unsigned long)name)) {
1361                 system->ref_count |= SYSTEM_FL_FREE_NAME;
1362                 system->name = kstrdup(name, GFP_KERNEL);
1363                 if (!system->name)
1364                         goto out_free;
1365         } else
1366                 system->name = name;
1367
1368         system->filter = NULL;
1369
1370         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1371         if (!system->filter)
1372                 goto out_free;
1373
1374         list_add(&system->list, &event_subsystems);
1375
1376         return system;
1377
1378  out_free:
1379         if (system->ref_count & SYSTEM_FL_FREE_NAME)
1380                 kfree(system->name);
1381         kfree(system);
1382         return NULL;
1383 }
1384
1385 static struct dentry *
1386 event_subsystem_dir(struct trace_array *tr, const char *name,
1387                     struct ftrace_event_file *file, struct dentry *parent)
1388 {
1389         struct ftrace_subsystem_dir *dir;
1390         struct event_subsystem *system;
1391         struct dentry *entry;
1392
1393         /* First see if we did not already create this dir */
1394         list_for_each_entry(dir, &tr->systems, list) {
1395                 system = dir->subsystem;
1396                 if (strcmp(system->name, name) == 0) {
1397                         dir->nr_events++;
1398                         file->system = dir;
1399                         return dir->entry;
1400                 }
1401         }
1402
1403         /* Now see if the system itself exists. */
1404         list_for_each_entry(system, &event_subsystems, list) {
1405                 if (strcmp(system->name, name) == 0)
1406                         break;
1407         }
1408         /* Reset system variable when not found */
1409         if (&system->list == &event_subsystems)
1410                 system = NULL;
1411
1412         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1413         if (!dir)
1414                 goto out_fail;
1415
1416         if (!system) {
1417                 system = create_new_subsystem(name);
1418                 if (!system)
1419                         goto out_free;
1420         } else
1421                 __get_system(system);
1422
1423         dir->entry = debugfs_create_dir(name, parent);
1424         if (!dir->entry) {
1425                 pr_warning("Failed to create system directory %s\n", name);
1426                 __put_system(system);
1427                 goto out_free;
1428         }
1429
1430         dir->tr = tr;
1431         dir->ref_count = 1;
1432         dir->nr_events = 1;
1433         dir->subsystem = system;
1434         file->system = dir;
1435
1436         entry = debugfs_create_file("filter", 0644, dir->entry, dir,
1437                                     &ftrace_subsystem_filter_fops);
1438         if (!entry) {
1439                 kfree(system->filter);
1440                 system->filter = NULL;
1441                 pr_warning("Could not create debugfs '%s/filter' entry\n", name);
1442         }
1443
1444         trace_create_file("enable", 0644, dir->entry, dir,
1445                           &ftrace_system_enable_fops);
1446
1447         list_add(&dir->list, &tr->systems);
1448
1449         return dir->entry;
1450
1451  out_free:
1452         kfree(dir);
1453  out_fail:
1454         /* Only print this message if failed on memory allocation */
1455         if (!dir || !system)
1456                 pr_warning("No memory to create event subsystem %s\n",
1457                            name);
1458         return NULL;
1459 }
1460
1461 static int
1462 event_create_dir(struct dentry *parent,
1463                  struct ftrace_event_file *file,
1464                  const struct file_operations *id,
1465                  const struct file_operations *enable,
1466                  const struct file_operations *filter,
1467                  const struct file_operations *format)
1468 {
1469         struct ftrace_event_call *call = file->event_call;
1470         struct trace_array *tr = file->tr;
1471         struct list_head *head;
1472         struct dentry *d_events;
1473         int ret;
1474
1475         /*
1476          * If the trace point header did not define TRACE_SYSTEM
1477          * then the system would be called "TRACE_SYSTEM".
1478          */
1479         if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1480                 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1481                 if (!d_events)
1482                         return -ENOMEM;
1483         } else
1484                 d_events = parent;
1485
1486         file->dir = debugfs_create_dir(call->name, d_events);
1487         if (!file->dir) {
1488                 pr_warning("Could not create debugfs '%s' directory\n",
1489                            call->name);
1490                 return -1;
1491         }
1492
1493         if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1494                 trace_create_file("enable", 0644, file->dir, file,
1495                                   enable);
1496
1497 #ifdef CONFIG_PERF_EVENTS
1498         if (call->event.type && call->class->reg)
1499                 trace_create_file("id", 0444, file->dir, call,
1500                                   id);
1501 #endif
1502
1503         /*
1504          * Other events may have the same class. Only update
1505          * the fields if they are not already defined.
1506          */
1507         head = trace_get_fields(call);
1508         if (list_empty(head)) {
1509                 ret = call->class->define_fields(call);
1510                 if (ret < 0) {
1511                         pr_warning("Could not initialize trace point"
1512                                    " events/%s\n", call->name);
1513                         return -1;
1514                 }
1515         }
1516         trace_create_file("filter", 0644, file->dir, call,
1517                           filter);
1518
1519         trace_create_file("format", 0444, file->dir, call,
1520                           format);
1521
1522         return 0;
1523 }
1524
1525 static void remove_subsystem(struct ftrace_subsystem_dir *dir)
1526 {
1527         if (!dir)
1528                 return;
1529
1530         if (!--dir->nr_events) {
1531                 debugfs_remove_recursive(dir->entry);
1532                 list_del(&dir->list);
1533                 __put_system_dir(dir);
1534         }
1535 }
1536
1537 static void remove_event_from_tracers(struct ftrace_event_call *call)
1538 {
1539         struct ftrace_event_file *file;
1540         struct trace_array *tr;
1541
1542         do_for_each_event_file_safe(tr, file) {
1543
1544                 if (file->event_call != call)
1545                         continue;
1546
1547                 list_del(&file->list);
1548                 debugfs_remove_recursive(file->dir);
1549                 remove_subsystem(file->system);
1550                 kmem_cache_free(file_cachep, file);
1551
1552                 /*
1553                  * The do_for_each_event_file_safe() is
1554                  * a double loop. After finding the call for this
1555                  * trace_array, we use break to jump to the next
1556                  * trace_array.
1557                  */
1558                 break;
1559         } while_for_each_event_file();
1560 }
1561
1562 static void event_remove(struct ftrace_event_call *call)
1563 {
1564         struct trace_array *tr;
1565         struct ftrace_event_file *file;
1566
1567         do_for_each_event_file(tr, file) {
1568                 if (file->event_call != call)
1569                         continue;
1570                 ftrace_event_enable_disable(file, 0);
1571                 /*
1572                  * The do_for_each_event_file() is
1573                  * a double loop. After finding the call for this
1574                  * trace_array, we use break to jump to the next
1575                  * trace_array.
1576                  */
1577                 break;
1578         } while_for_each_event_file();
1579
1580         if (call->event.funcs)
1581                 __unregister_ftrace_event(&call->event);
1582         remove_event_from_tracers(call);
1583         list_del(&call->list);
1584 }
1585
1586 static int event_init(struct ftrace_event_call *call)
1587 {
1588         int ret = 0;
1589
1590         if (WARN_ON(!call->name))
1591                 return -EINVAL;
1592
1593         if (call->class->raw_init) {
1594                 ret = call->class->raw_init(call);
1595                 if (ret < 0 && ret != -ENOSYS)
1596                         pr_warn("Could not initialize trace events/%s\n",
1597                                 call->name);
1598         }
1599
1600         return ret;
1601 }
1602
1603 static int
1604 __register_event(struct ftrace_event_call *call, struct module *mod)
1605 {
1606         int ret;
1607
1608         ret = event_init(call);
1609         if (ret < 0)
1610                 return ret;
1611
1612         list_add(&call->list, &ftrace_events);
1613         call->mod = mod;
1614
1615         return 0;
1616 }
1617
1618 static struct ftrace_event_file *
1619 trace_create_new_event(struct ftrace_event_call *call,
1620                        struct trace_array *tr)
1621 {
1622         struct ftrace_event_file *file;
1623
1624         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
1625         if (!file)
1626                 return NULL;
1627
1628         file->event_call = call;
1629         file->tr = tr;
1630         atomic_set(&file->sm_ref, 0);
1631         list_add(&file->list, &tr->events);
1632
1633         return file;
1634 }
1635
1636 /* Add an event to a trace directory */
1637 static int
1638 __trace_add_new_event(struct ftrace_event_call *call,
1639                       struct trace_array *tr,
1640                       const struct file_operations *id,
1641                       const struct file_operations *enable,
1642                       const struct file_operations *filter,
1643                       const struct file_operations *format)
1644 {
1645         struct ftrace_event_file *file;
1646
1647         file = trace_create_new_event(call, tr);
1648         if (!file)
1649                 return -ENOMEM;
1650
1651         return event_create_dir(tr->event_dir, file, id, enable, filter, format);
1652 }
1653
1654 /*
1655  * Just create a decriptor for early init. A descriptor is required
1656  * for enabling events at boot. We want to enable events before
1657  * the filesystem is initialized.
1658  */
1659 static __init int
1660 __trace_early_add_new_event(struct ftrace_event_call *call,
1661                             struct trace_array *tr)
1662 {
1663         struct ftrace_event_file *file;
1664
1665         file = trace_create_new_event(call, tr);
1666         if (!file)
1667                 return -ENOMEM;
1668
1669         return 0;
1670 }
1671
1672 struct ftrace_module_file_ops;
1673 static void __add_event_to_tracers(struct ftrace_event_call *call,
1674                                    struct ftrace_module_file_ops *file_ops);
1675
1676 /* Add an additional event_call dynamically */
1677 int trace_add_event_call(struct ftrace_event_call *call)
1678 {
1679         int ret;
1680         mutex_lock(&trace_types_lock);
1681         mutex_lock(&event_mutex);
1682
1683         ret = __register_event(call, NULL);
1684         if (ret >= 0)
1685                 __add_event_to_tracers(call, NULL);
1686
1687         mutex_unlock(&event_mutex);
1688         mutex_unlock(&trace_types_lock);
1689         return ret;
1690 }
1691
1692 /*
1693  * Must be called under locking of trace_types_lock, event_mutex and
1694  * trace_event_sem.
1695  */
1696 static void __trace_remove_event_call(struct ftrace_event_call *call)
1697 {
1698         event_remove(call);
1699         trace_destroy_fields(call);
1700         destroy_preds(call);
1701 }
1702
1703 /* Remove an event_call */
1704 void trace_remove_event_call(struct ftrace_event_call *call)
1705 {
1706         mutex_lock(&trace_types_lock);
1707         mutex_lock(&event_mutex);
1708         down_write(&trace_event_sem);
1709         __trace_remove_event_call(call);
1710         up_write(&trace_event_sem);
1711         mutex_unlock(&event_mutex);
1712         mutex_unlock(&trace_types_lock);
1713 }
1714
1715 #define for_each_event(event, start, end)                       \
1716         for (event = start;                                     \
1717              (unsigned long)event < (unsigned long)end;         \
1718              event++)
1719
1720 #ifdef CONFIG_MODULES
1721
1722 static LIST_HEAD(ftrace_module_file_list);
1723
1724 /*
1725  * Modules must own their file_operations to keep up with
1726  * reference counting.
1727  */
1728 struct ftrace_module_file_ops {
1729         struct list_head                list;
1730         struct module                   *mod;
1731         struct file_operations          id;
1732         struct file_operations          enable;
1733         struct file_operations          format;
1734         struct file_operations          filter;
1735 };
1736
1737 static struct ftrace_module_file_ops *
1738 find_ftrace_file_ops(struct ftrace_module_file_ops *file_ops, struct module *mod)
1739 {
1740         /*
1741          * As event_calls are added in groups by module,
1742          * when we find one file_ops, we don't need to search for
1743          * each call in that module, as the rest should be the
1744          * same. Only search for a new one if the last one did
1745          * not match.
1746          */
1747         if (file_ops && mod == file_ops->mod)
1748                 return file_ops;
1749
1750         list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1751                 if (file_ops->mod == mod)
1752                         return file_ops;
1753         }
1754         return NULL;
1755 }
1756
1757 static struct ftrace_module_file_ops *
1758 trace_create_file_ops(struct module *mod)
1759 {
1760         struct ftrace_module_file_ops *file_ops;
1761
1762         /*
1763          * This is a bit of a PITA. To allow for correct reference
1764          * counting, modules must "own" their file_operations.
1765          * To do this, we allocate the file operations that will be
1766          * used in the event directory.
1767          */
1768
1769         file_ops = kmalloc(sizeof(*file_ops), GFP_KERNEL);
1770         if (!file_ops)
1771                 return NULL;
1772
1773         file_ops->mod = mod;
1774
1775         file_ops->id = ftrace_event_id_fops;
1776         file_ops->id.owner = mod;
1777
1778         file_ops->enable = ftrace_enable_fops;
1779         file_ops->enable.owner = mod;
1780
1781         file_ops->filter = ftrace_event_filter_fops;
1782         file_ops->filter.owner = mod;
1783
1784         file_ops->format = ftrace_event_format_fops;
1785         file_ops->format.owner = mod;
1786
1787         list_add(&file_ops->list, &ftrace_module_file_list);
1788
1789         return file_ops;
1790 }
1791
1792 static void trace_module_add_events(struct module *mod)
1793 {
1794         struct ftrace_module_file_ops *file_ops = NULL;
1795         struct ftrace_event_call **call, **start, **end;
1796
1797         start = mod->trace_events;
1798         end = mod->trace_events + mod->num_trace_events;
1799
1800         if (start == end)
1801                 return;
1802
1803         file_ops = trace_create_file_ops(mod);
1804         if (!file_ops)
1805                 return;
1806
1807         for_each_event(call, start, end) {
1808                 __register_event(*call, mod);
1809                 __add_event_to_tracers(*call, file_ops);
1810         }
1811 }
1812
1813 static void trace_module_remove_events(struct module *mod)
1814 {
1815         struct ftrace_module_file_ops *file_ops;
1816         struct ftrace_event_call *call, *p;
1817         bool clear_trace = false;
1818
1819         down_write(&trace_event_sem);
1820         list_for_each_entry_safe(call, p, &ftrace_events, list) {
1821                 if (call->mod == mod) {
1822                         if (call->flags & TRACE_EVENT_FL_WAS_ENABLED)
1823                                 clear_trace = true;
1824                         __trace_remove_event_call(call);
1825                 }
1826         }
1827
1828         /* Now free the file_operations */
1829         list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1830                 if (file_ops->mod == mod)
1831                         break;
1832         }
1833         if (&file_ops->list != &ftrace_module_file_list) {
1834                 list_del(&file_ops->list);
1835                 kfree(file_ops);
1836         }
1837         up_write(&trace_event_sem);
1838
1839         /*
1840          * It is safest to reset the ring buffer if the module being unloaded
1841          * registered any events that were used. The only worry is if
1842          * a new module gets loaded, and takes on the same id as the events
1843          * of this module. When printing out the buffer, traced events left
1844          * over from this module may be passed to the new module events and
1845          * unexpected results may occur.
1846          */
1847         if (clear_trace)
1848                 tracing_reset_all_online_cpus();
1849 }
1850
1851 static int trace_module_notify(struct notifier_block *self,
1852                                unsigned long val, void *data)
1853 {
1854         struct module *mod = data;
1855
1856         mutex_lock(&trace_types_lock);
1857         mutex_lock(&event_mutex);
1858         switch (val) {
1859         case MODULE_STATE_COMING:
1860                 trace_module_add_events(mod);
1861                 break;
1862         case MODULE_STATE_GOING:
1863                 trace_module_remove_events(mod);
1864                 break;
1865         }
1866         mutex_unlock(&event_mutex);
1867         mutex_unlock(&trace_types_lock);
1868
1869         return 0;
1870 }
1871
1872 static int
1873 __trace_add_new_mod_event(struct ftrace_event_call *call,
1874                           struct trace_array *tr,
1875                           struct ftrace_module_file_ops *file_ops)
1876 {
1877         return __trace_add_new_event(call, tr,
1878                                      &file_ops->id, &file_ops->enable,
1879                                      &file_ops->filter, &file_ops->format);
1880 }
1881
1882 #else
1883 static inline struct ftrace_module_file_ops *
1884 find_ftrace_file_ops(struct ftrace_module_file_ops *file_ops, struct module *mod)
1885 {
1886         return NULL;
1887 }
1888 static inline int trace_module_notify(struct notifier_block *self,
1889                                       unsigned long val, void *data)
1890 {
1891         return 0;
1892 }
1893 static inline int
1894 __trace_add_new_mod_event(struct ftrace_event_call *call,
1895                           struct trace_array *tr,
1896                           struct ftrace_module_file_ops *file_ops)
1897 {
1898         return -ENODEV;
1899 }
1900 #endif /* CONFIG_MODULES */
1901
1902 /* Create a new event directory structure for a trace directory. */
1903 static void
1904 __trace_add_event_dirs(struct trace_array *tr)
1905 {
1906         struct ftrace_module_file_ops *file_ops = NULL;
1907         struct ftrace_event_call *call;
1908         int ret;
1909
1910         list_for_each_entry(call, &ftrace_events, list) {
1911                 if (call->mod) {
1912                         /*
1913                          * Directories for events by modules need to
1914                          * keep module ref counts when opened (as we don't
1915                          * want the module to disappear when reading one
1916                          * of these files). The file_ops keep account of
1917                          * the module ref count.
1918                          */
1919                         file_ops = find_ftrace_file_ops(file_ops, call->mod);
1920                         if (!file_ops)
1921                                 continue; /* Warn? */
1922                         ret = __trace_add_new_mod_event(call, tr, file_ops);
1923                         if (ret < 0)
1924                                 pr_warning("Could not create directory for event %s\n",
1925                                            call->name);
1926                         continue;
1927                 }
1928                 ret = __trace_add_new_event(call, tr,
1929                                             &ftrace_event_id_fops,
1930                                             &ftrace_enable_fops,
1931                                             &ftrace_event_filter_fops,
1932                                             &ftrace_event_format_fops);
1933                 if (ret < 0)
1934                         pr_warning("Could not create directory for event %s\n",
1935                                    call->name);
1936         }
1937 }
1938
1939 #ifdef CONFIG_DYNAMIC_FTRACE
1940
1941 /* Avoid typos */
1942 #define ENABLE_EVENT_STR        "enable_event"
1943 #define DISABLE_EVENT_STR       "disable_event"
1944
1945 struct event_probe_data {
1946         struct ftrace_event_file        *file;
1947         unsigned long                   count;
1948         int                             ref;
1949         bool                            enable;
1950 };
1951
1952 static struct ftrace_event_file *
1953 find_event_file(struct trace_array *tr, const char *system,  const char *event)
1954 {
1955         struct ftrace_event_file *file;
1956         struct ftrace_event_call *call;
1957
1958         list_for_each_entry(file, &tr->events, list) {
1959
1960                 call = file->event_call;
1961
1962                 if (!call->name || !call->class || !call->class->reg)
1963                         continue;
1964
1965                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
1966                         continue;
1967
1968                 if (strcmp(event, call->name) == 0 &&
1969                     strcmp(system, call->class->system) == 0)
1970                         return file;
1971         }
1972         return NULL;
1973 }
1974
1975 static void
1976 event_enable_probe(unsigned long ip, unsigned long parent_ip, void **_data)
1977 {
1978         struct event_probe_data **pdata = (struct event_probe_data **)_data;
1979         struct event_probe_data *data = *pdata;
1980
1981         if (!data)
1982                 return;
1983
1984         if (data->enable)
1985                 clear_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &data->file->flags);
1986         else
1987                 set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &data->file->flags);
1988 }
1989
1990 static void
1991 event_enable_count_probe(unsigned long ip, unsigned long parent_ip, void **_data)
1992 {
1993         struct event_probe_data **pdata = (struct event_probe_data **)_data;
1994         struct event_probe_data *data = *pdata;
1995
1996         if (!data)
1997                 return;
1998
1999         if (!data->count)
2000                 return;
2001
2002         /* Skip if the event is in a state we want to switch to */
2003         if (data->enable == !(data->file->flags & FTRACE_EVENT_FL_SOFT_DISABLED))
2004                 return;
2005
2006         if (data->count != -1)
2007                 (data->count)--;
2008
2009         event_enable_probe(ip, parent_ip, _data);
2010 }
2011
2012 static int
2013 event_enable_print(struct seq_file *m, unsigned long ip,
2014                       struct ftrace_probe_ops *ops, void *_data)
2015 {
2016         struct event_probe_data *data = _data;
2017
2018         seq_printf(m, "%ps:", (void *)ip);
2019
2020         seq_printf(m, "%s:%s:%s",
2021                    data->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2022                    data->file->event_call->class->system,
2023                    data->file->event_call->name);
2024
2025         if (data->count == -1)
2026                 seq_printf(m, ":unlimited\n");
2027         else
2028                 seq_printf(m, ":count=%ld\n", data->count);
2029
2030         return 0;
2031 }
2032
2033 static int
2034 event_enable_init(struct ftrace_probe_ops *ops, unsigned long ip,
2035                   void **_data)
2036 {
2037         struct event_probe_data **pdata = (struct event_probe_data **)_data;
2038         struct event_probe_data *data = *pdata;
2039
2040         data->ref++;
2041         return 0;
2042 }
2043
2044 static void
2045 event_enable_free(struct ftrace_probe_ops *ops, unsigned long ip,
2046                   void **_data)
2047 {
2048         struct event_probe_data **pdata = (struct event_probe_data **)_data;
2049         struct event_probe_data *data = *pdata;
2050
2051         if (WARN_ON_ONCE(data->ref <= 0))
2052                 return;
2053
2054         data->ref--;
2055         if (!data->ref) {
2056                 /* Remove the SOFT_MODE flag */
2057                 __ftrace_event_enable_disable(data->file, 0, 1);
2058                 module_put(data->file->event_call->mod);
2059                 kfree(data);
2060         }
2061         *pdata = NULL;
2062 }
2063
2064 static struct ftrace_probe_ops event_enable_probe_ops = {
2065         .func                   = event_enable_probe,
2066         .print                  = event_enable_print,
2067         .init                   = event_enable_init,
2068         .free                   = event_enable_free,
2069 };
2070
2071 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2072         .func                   = event_enable_count_probe,
2073         .print                  = event_enable_print,
2074         .init                   = event_enable_init,
2075         .free                   = event_enable_free,
2076 };
2077
2078 static struct ftrace_probe_ops event_disable_probe_ops = {
2079         .func                   = event_enable_probe,
2080         .print                  = event_enable_print,
2081         .init                   = event_enable_init,
2082         .free                   = event_enable_free,
2083 };
2084
2085 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2086         .func                   = event_enable_count_probe,
2087         .print                  = event_enable_print,
2088         .init                   = event_enable_init,
2089         .free                   = event_enable_free,
2090 };
2091
2092 static int
2093 event_enable_func(struct ftrace_hash *hash,
2094                   char *glob, char *cmd, char *param, int enabled)
2095 {
2096         struct trace_array *tr = top_trace_array();
2097         struct ftrace_event_file *file;
2098         struct ftrace_probe_ops *ops;
2099         struct event_probe_data *data;
2100         const char *system;
2101         const char *event;
2102         char *number;
2103         bool enable;
2104         int ret;
2105
2106         /* hash funcs only work with set_ftrace_filter */
2107         if (!enabled || !param)
2108                 return -EINVAL;
2109
2110         system = strsep(&param, ":");
2111         if (!param)
2112                 return -EINVAL;
2113
2114         event = strsep(&param, ":");
2115
2116         mutex_lock(&event_mutex);
2117
2118         ret = -EINVAL;
2119         file = find_event_file(tr, system, event);
2120         if (!file)
2121                 goto out;
2122
2123         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2124
2125         if (enable)
2126                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2127         else
2128                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2129
2130         if (glob[0] == '!') {
2131                 unregister_ftrace_function_probe_func(glob+1, ops);
2132                 ret = 0;
2133                 goto out;
2134         }
2135
2136         ret = -ENOMEM;
2137         data = kzalloc(sizeof(*data), GFP_KERNEL);
2138         if (!data)
2139                 goto out;
2140
2141         data->enable = enable;
2142         data->count = -1;
2143         data->file = file;
2144
2145         if (!param)
2146                 goto out_reg;
2147
2148         number = strsep(&param, ":");
2149
2150         ret = -EINVAL;
2151         if (!strlen(number))
2152                 goto out_free;
2153
2154         /*
2155          * We use the callback data field (which is a pointer)
2156          * as our counter.
2157          */
2158         ret = kstrtoul(number, 0, &data->count);
2159         if (ret)
2160                 goto out_free;
2161
2162  out_reg:
2163         /* Don't let event modules unload while probe registered */
2164         ret = try_module_get(file->event_call->mod);
2165         if (!ret) {
2166                 ret = -EBUSY;
2167                 goto out_free;
2168         }
2169
2170         ret = __ftrace_event_enable_disable(file, 1, 1);
2171         if (ret < 0)
2172                 goto out_put;
2173         ret = register_ftrace_function_probe(glob, ops, data);
2174         /*
2175          * The above returns on success the # of functions enabled,
2176          * but if it didn't find any functions it returns zero.
2177          * Consider no functions a failure too.
2178          */
2179         if (!ret) {
2180                 ret = -ENOENT;
2181                 goto out_disable;
2182         } else if (ret < 0)
2183                 goto out_disable;
2184         /* Just return zero, not the number of enabled functions */
2185         ret = 0;
2186  out:
2187         mutex_unlock(&event_mutex);
2188         return ret;
2189
2190  out_disable:
2191         __ftrace_event_enable_disable(file, 0, 1);
2192  out_put:
2193         module_put(file->event_call->mod);
2194  out_free:
2195         kfree(data);
2196         goto out;
2197 }
2198
2199 static struct ftrace_func_command event_enable_cmd = {
2200         .name                   = ENABLE_EVENT_STR,
2201         .func                   = event_enable_func,
2202 };
2203
2204 static struct ftrace_func_command event_disable_cmd = {
2205         .name                   = DISABLE_EVENT_STR,
2206         .func                   = event_enable_func,
2207 };
2208
2209 static __init int register_event_cmds(void)
2210 {
2211         int ret;
2212
2213         ret = register_ftrace_command(&event_enable_cmd);
2214         if (WARN_ON(ret < 0))
2215                 return ret;
2216         ret = register_ftrace_command(&event_disable_cmd);
2217         if (WARN_ON(ret < 0))
2218                 unregister_ftrace_command(&event_enable_cmd);
2219         return ret;
2220 }
2221 #else
2222 static inline int register_event_cmds(void) { return 0; }
2223 #endif /* CONFIG_DYNAMIC_FTRACE */
2224
2225 /*
2226  * The top level array has already had its ftrace_event_file
2227  * descriptors created in order to allow for early events to
2228  * be recorded. This function is called after the debugfs has been
2229  * initialized, and we now have to create the files associated
2230  * to the events.
2231  */
2232 static __init void
2233 __trace_early_add_event_dirs(struct trace_array *tr)
2234 {
2235         struct ftrace_event_file *file;
2236         int ret;
2237
2238
2239         list_for_each_entry(file, &tr->events, list) {
2240                 ret = event_create_dir(tr->event_dir, file,
2241                                        &ftrace_event_id_fops,
2242                                        &ftrace_enable_fops,
2243                                        &ftrace_event_filter_fops,
2244                                        &ftrace_event_format_fops);
2245                 if (ret < 0)
2246                         pr_warning("Could not create directory for event %s\n",
2247                                    file->event_call->name);
2248         }
2249 }
2250
2251 /*
2252  * For early boot up, the top trace array requires to have
2253  * a list of events that can be enabled. This must be done before
2254  * the filesystem is set up in order to allow events to be traced
2255  * early.
2256  */
2257 static __init void
2258 __trace_early_add_events(struct trace_array *tr)
2259 {
2260         struct ftrace_event_call *call;
2261         int ret;
2262
2263         list_for_each_entry(call, &ftrace_events, list) {
2264                 /* Early boot up should not have any modules loaded */
2265                 if (WARN_ON_ONCE(call->mod))
2266                         continue;
2267
2268                 ret = __trace_early_add_new_event(call, tr);
2269                 if (ret < 0)
2270                         pr_warning("Could not create early event %s\n",
2271                                    call->name);
2272         }
2273 }
2274
2275 /* Remove the event directory structure for a trace directory. */
2276 static void
2277 __trace_remove_event_dirs(struct trace_array *tr)
2278 {
2279         struct ftrace_event_file *file, *next;
2280
2281         list_for_each_entry_safe(file, next, &tr->events, list) {
2282                 list_del(&file->list);
2283                 debugfs_remove_recursive(file->dir);
2284                 remove_subsystem(file->system);
2285                 kmem_cache_free(file_cachep, file);
2286         }
2287 }
2288
2289 static void
2290 __add_event_to_tracers(struct ftrace_event_call *call,
2291                        struct ftrace_module_file_ops *file_ops)
2292 {
2293         struct trace_array *tr;
2294
2295         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2296                 if (file_ops)
2297                         __trace_add_new_mod_event(call, tr, file_ops);
2298                 else
2299                         __trace_add_new_event(call, tr,
2300                                               &ftrace_event_id_fops,
2301                                               &ftrace_enable_fops,
2302                                               &ftrace_event_filter_fops,
2303                                               &ftrace_event_format_fops);
2304         }
2305 }
2306
2307 static struct notifier_block trace_module_nb = {
2308         .notifier_call = trace_module_notify,
2309         .priority = 0,
2310 };
2311
2312 extern struct ftrace_event_call *__start_ftrace_events[];
2313 extern struct ftrace_event_call *__stop_ftrace_events[];
2314
2315 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2316
2317 static __init int setup_trace_event(char *str)
2318 {
2319         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2320         ring_buffer_expanded = true;
2321         tracing_selftest_disabled = true;
2322
2323         return 1;
2324 }
2325 __setup("trace_event=", setup_trace_event);
2326
2327 /* Expects to have event_mutex held when called */
2328 static int
2329 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2330 {
2331         struct dentry *d_events;
2332         struct dentry *entry;
2333
2334         entry = debugfs_create_file("set_event", 0644, parent,
2335                                     tr, &ftrace_set_event_fops);
2336         if (!entry) {
2337                 pr_warning("Could not create debugfs 'set_event' entry\n");
2338                 return -ENOMEM;
2339         }
2340
2341         d_events = debugfs_create_dir("events", parent);
2342         if (!d_events) {
2343                 pr_warning("Could not create debugfs 'events' directory\n");
2344                 return -ENOMEM;
2345         }
2346
2347         /* ring buffer internal formats */
2348         trace_create_file("header_page", 0444, d_events,
2349                           ring_buffer_print_page_header,
2350                           &ftrace_show_header_fops);
2351
2352         trace_create_file("header_event", 0444, d_events,
2353                           ring_buffer_print_entry_header,
2354                           &ftrace_show_header_fops);
2355
2356         trace_create_file("enable", 0644, d_events,
2357                           tr, &ftrace_tr_enable_fops);
2358
2359         tr->event_dir = d_events;
2360
2361         return 0;
2362 }
2363
2364 /**
2365  * event_trace_add_tracer - add a instance of a trace_array to events
2366  * @parent: The parent dentry to place the files/directories for events in
2367  * @tr: The trace array associated with these events
2368  *
2369  * When a new instance is created, it needs to set up its events
2370  * directory, as well as other files associated with events. It also
2371  * creates the event hierachry in the @parent/events directory.
2372  *
2373  * Returns 0 on success.
2374  */
2375 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2376 {
2377         int ret;
2378
2379         mutex_lock(&event_mutex);
2380
2381         ret = create_event_toplevel_files(parent, tr);
2382         if (ret)
2383                 goto out_unlock;
2384
2385         down_write(&trace_event_sem);
2386         __trace_add_event_dirs(tr);
2387         up_write(&trace_event_sem);
2388
2389  out_unlock:
2390         mutex_unlock(&event_mutex);
2391
2392         return ret;
2393 }
2394
2395 /*
2396  * The top trace array already had its file descriptors created.
2397  * Now the files themselves need to be created.
2398  */
2399 static __init int
2400 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
2401 {
2402         int ret;
2403
2404         mutex_lock(&event_mutex);
2405
2406         ret = create_event_toplevel_files(parent, tr);
2407         if (ret)
2408                 goto out_unlock;
2409
2410         down_write(&trace_event_sem);
2411         __trace_early_add_event_dirs(tr);
2412         up_write(&trace_event_sem);
2413
2414  out_unlock:
2415         mutex_unlock(&event_mutex);
2416
2417         return ret;
2418 }
2419
2420 int event_trace_del_tracer(struct trace_array *tr)
2421 {
2422         mutex_lock(&event_mutex);
2423
2424         /* Disable any running events */
2425         __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
2426
2427         down_write(&trace_event_sem);
2428         __trace_remove_event_dirs(tr);
2429         debugfs_remove_recursive(tr->event_dir);
2430         up_write(&trace_event_sem);
2431
2432         tr->event_dir = NULL;
2433
2434         mutex_unlock(&event_mutex);
2435
2436         return 0;
2437 }
2438
2439 static __init int event_trace_memsetup(void)
2440 {
2441         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
2442         file_cachep = KMEM_CACHE(ftrace_event_file, SLAB_PANIC);
2443         return 0;
2444 }
2445
2446 static __init int event_trace_enable(void)
2447 {
2448         struct trace_array *tr = top_trace_array();
2449         struct ftrace_event_call **iter, *call;
2450         char *buf = bootup_event_buf;
2451         char *token;
2452         int ret;
2453
2454         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
2455
2456                 call = *iter;
2457                 ret = event_init(call);
2458                 if (!ret)
2459                         list_add(&call->list, &ftrace_events);
2460         }
2461
2462         /*
2463          * We need the top trace array to have a working set of trace
2464          * points at early init, before the debug files and directories
2465          * are created. Create the file entries now, and attach them
2466          * to the actual file dentries later.
2467          */
2468         __trace_early_add_events(tr);
2469
2470         while (true) {
2471                 token = strsep(&buf, ",");
2472
2473                 if (!token)
2474                         break;
2475                 if (!*token)
2476                         continue;
2477
2478                 ret = ftrace_set_clr_event(tr, token, 1);
2479                 if (ret)
2480                         pr_warn("Failed to enable trace event: %s\n", token);
2481         }
2482
2483         trace_printk_start_comm();
2484
2485         register_event_cmds();
2486
2487         return 0;
2488 }
2489
2490 static __init int event_trace_init(void)
2491 {
2492         struct trace_array *tr;
2493         struct dentry *d_tracer;
2494         struct dentry *entry;
2495         int ret;
2496
2497         tr = top_trace_array();
2498
2499         d_tracer = tracing_init_dentry();
2500         if (!d_tracer)
2501                 return 0;
2502
2503         entry = debugfs_create_file("available_events", 0444, d_tracer,
2504                                     tr, &ftrace_avail_fops);
2505         if (!entry)
2506                 pr_warning("Could not create debugfs "
2507                            "'available_events' entry\n");
2508
2509         if (trace_define_common_fields())
2510                 pr_warning("tracing: Failed to allocate common fields");
2511
2512         ret = early_event_add_tracer(d_tracer, tr);
2513         if (ret)
2514                 return ret;
2515
2516         ret = register_module_notifier(&trace_module_nb);
2517         if (ret)
2518                 pr_warning("Failed to register trace events module notifier\n");
2519
2520         return 0;
2521 }
2522 early_initcall(event_trace_memsetup);
2523 core_initcall(event_trace_enable);
2524 fs_initcall(event_trace_init);
2525
2526 #ifdef CONFIG_FTRACE_STARTUP_TEST
2527
2528 static DEFINE_SPINLOCK(test_spinlock);
2529 static DEFINE_SPINLOCK(test_spinlock_irq);
2530 static DEFINE_MUTEX(test_mutex);
2531
2532 static __init void test_work(struct work_struct *dummy)
2533 {
2534         spin_lock(&test_spinlock);
2535         spin_lock_irq(&test_spinlock_irq);
2536         udelay(1);
2537         spin_unlock_irq(&test_spinlock_irq);
2538         spin_unlock(&test_spinlock);
2539
2540         mutex_lock(&test_mutex);
2541         msleep(1);
2542         mutex_unlock(&test_mutex);
2543 }
2544
2545 static __init int event_test_thread(void *unused)
2546 {
2547         void *test_malloc;
2548
2549         test_malloc = kmalloc(1234, GFP_KERNEL);
2550         if (!test_malloc)
2551                 pr_info("failed to kmalloc\n");
2552
2553         schedule_on_each_cpu(test_work);
2554
2555         kfree(test_malloc);
2556
2557         set_current_state(TASK_INTERRUPTIBLE);
2558         while (!kthread_should_stop())
2559                 schedule();
2560
2561         return 0;
2562 }
2563
2564 /*
2565  * Do various things that may trigger events.
2566  */
2567 static __init void event_test_stuff(void)
2568 {
2569         struct task_struct *test_thread;
2570
2571         test_thread = kthread_run(event_test_thread, NULL, "test-events");
2572         msleep(1);
2573         kthread_stop(test_thread);
2574 }
2575
2576 /*
2577  * For every trace event defined, we will test each trace point separately,
2578  * and then by groups, and finally all trace points.
2579  */
2580 static __init void event_trace_self_tests(void)
2581 {
2582         struct ftrace_subsystem_dir *dir;
2583         struct ftrace_event_file *file;
2584         struct ftrace_event_call *call;
2585         struct event_subsystem *system;
2586         struct trace_array *tr;
2587         int ret;
2588
2589         tr = top_trace_array();
2590
2591         pr_info("Running tests on trace events:\n");
2592
2593         list_for_each_entry(file, &tr->events, list) {
2594
2595                 call = file->event_call;
2596
2597                 /* Only test those that have a probe */
2598                 if (!call->class || !call->class->probe)
2599                         continue;
2600
2601 /*
2602  * Testing syscall events here is pretty useless, but
2603  * we still do it if configured. But this is time consuming.
2604  * What we really need is a user thread to perform the
2605  * syscalls as we test.
2606  */
2607 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
2608                 if (call->class->system &&
2609                     strcmp(call->class->system, "syscalls") == 0)
2610                         continue;
2611 #endif
2612
2613                 pr_info("Testing event %s: ", call->name);
2614
2615                 /*
2616                  * If an event is already enabled, someone is using
2617                  * it and the self test should not be on.
2618                  */
2619                 if (file->flags & FTRACE_EVENT_FL_ENABLED) {
2620                         pr_warning("Enabled event during self test!\n");
2621                         WARN_ON_ONCE(1);
2622                         continue;
2623                 }
2624
2625                 ftrace_event_enable_disable(file, 1);
2626                 event_test_stuff();
2627                 ftrace_event_enable_disable(file, 0);
2628
2629                 pr_cont("OK\n");
2630         }
2631
2632         /* Now test at the sub system level */
2633
2634         pr_info("Running tests on trace event systems:\n");
2635
2636         list_for_each_entry(dir, &tr->systems, list) {
2637
2638                 system = dir->subsystem;
2639
2640                 /* the ftrace system is special, skip it */
2641                 if (strcmp(system->name, "ftrace") == 0)
2642                         continue;
2643
2644                 pr_info("Testing event system %s: ", system->name);
2645
2646                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
2647                 if (WARN_ON_ONCE(ret)) {
2648                         pr_warning("error enabling system %s\n",
2649                                    system->name);
2650                         continue;
2651                 }
2652
2653                 event_test_stuff();
2654
2655                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
2656                 if (WARN_ON_ONCE(ret)) {
2657                         pr_warning("error disabling system %s\n",
2658                                    system->name);
2659                         continue;
2660                 }
2661
2662                 pr_cont("OK\n");
2663         }
2664
2665         /* Test with all events enabled */
2666
2667         pr_info("Running tests on all trace events:\n");
2668         pr_info("Testing all events: ");
2669
2670         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
2671         if (WARN_ON_ONCE(ret)) {
2672                 pr_warning("error enabling all events\n");
2673                 return;
2674         }
2675
2676         event_test_stuff();
2677
2678         /* reset sysname */
2679         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
2680         if (WARN_ON_ONCE(ret)) {
2681                 pr_warning("error disabling all events\n");
2682                 return;
2683         }
2684
2685         pr_cont("OK\n");
2686 }
2687
2688 #ifdef CONFIG_FUNCTION_TRACER
2689
2690 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
2691
2692 static void
2693 function_test_events_call(unsigned long ip, unsigned long parent_ip,
2694                           struct ftrace_ops *op, struct pt_regs *pt_regs)
2695 {
2696         struct ring_buffer_event *event;
2697         struct ring_buffer *buffer;
2698         struct ftrace_entry *entry;
2699         unsigned long flags;
2700         long disabled;
2701         int cpu;
2702         int pc;
2703
2704         pc = preempt_count();
2705         preempt_disable_notrace();
2706         cpu = raw_smp_processor_id();
2707         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
2708
2709         if (disabled != 1)
2710                 goto out;
2711
2712         local_save_flags(flags);
2713
2714         event = trace_current_buffer_lock_reserve(&buffer,
2715                                                   TRACE_FN, sizeof(*entry),
2716                                                   flags, pc);
2717         if (!event)
2718                 goto out;
2719         entry   = ring_buffer_event_data(event);
2720         entry->ip                       = ip;
2721         entry->parent_ip                = parent_ip;
2722
2723         trace_buffer_unlock_commit(buffer, event, flags, pc);
2724
2725  out:
2726         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
2727         preempt_enable_notrace();
2728 }
2729
2730 static struct ftrace_ops trace_ops __initdata  =
2731 {
2732         .func = function_test_events_call,
2733         .flags = FTRACE_OPS_FL_RECURSION_SAFE,
2734 };
2735
2736 static __init void event_trace_self_test_with_function(void)
2737 {
2738         int ret;
2739         ret = register_ftrace_function(&trace_ops);
2740         if (WARN_ON(ret < 0)) {
2741                 pr_info("Failed to enable function tracer for event tests\n");
2742                 return;
2743         }
2744         pr_info("Running tests again, along with the function tracer\n");
2745         event_trace_self_tests();
2746         unregister_ftrace_function(&trace_ops);
2747 }
2748 #else
2749 static __init void event_trace_self_test_with_function(void)
2750 {
2751 }
2752 #endif
2753
2754 static __init int event_trace_self_tests_init(void)
2755 {
2756         if (!tracing_selftest_disabled) {
2757                 event_trace_self_tests();
2758                 event_trace_self_test_with_function();
2759         }
2760
2761         return 0;
2762 }
2763
2764 late_initcall(event_trace_self_tests_init);
2765
2766 #endif