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/delay.h>
  19
  20 #include <asm/setup.h>
  21
  22 #include "trace_output.h"
  23
  24 #undef TRACE_SYSTEM
  25 #define TRACE_SYSTEM "TRACE_SYSTEM"
  26
  27 DEFINE_MUTEX(event_mutex);
  28
  29 LIST_HEAD(ftrace_events);
  30
  31 int trace_define_field(struct ftrace_event_call *call, const char *type,
  32                        const char *name, int offset, int size, int is_signed,
  33                        int filter_type)
  34 {
  35         struct ftrace_event_field *field;
  36
  37         field = kzalloc(sizeof(*field), GFP_KERNEL);
  38         if (!field)
  39                 goto err;
  40
  41         field->name = kstrdup(name, GFP_KERNEL);
  42         if (!field->name)
  43                 goto err;
  44
  45         field->type = kstrdup(type, GFP_KERNEL);
  46         if (!field->type)
  47                 goto err;
  48
  49         if (filter_type == FILTER_OTHER)
  50                 field->filter_type = filter_assign_type(type);
  51         else
  52                 field->filter_type = filter_type;
  53
  54         field->offset = offset;
  55         field->size = size;
  56         field->is_signed = is_signed;
  57
  58         list_add(&field->link, &call->fields);
  59
  60         return 0;
  61
  62 err:
  63         if (field) {
  64                 kfree(field->name);
  65                 kfree(field->type);
  66         }
  67         kfree(field);
  68
  69         return -ENOMEM;
  70 }
  71 EXPORT_SYMBOL_GPL(trace_define_field);
  72
  73 #define __common_field(type, item)                                      \
  74         ret = trace_define_field(call, #type, "common_" #item,          \
  75                                  offsetof(typeof(ent), item),           \
  76                                  sizeof(ent.item),                      \
  77                                  is_signed_type(type), FILTER_OTHER);   \
  78         if (ret)                                                        \
  79                 return ret;
  80
  81 int trace_define_common_fields(struct ftrace_event_call *call)
  82 {
  83         int ret;
  84         struct trace_entry ent;
  85
  86         __common_field(unsigned short, type);
  87         __common_field(unsigned char, flags);
  88         __common_field(unsigned char, preempt_count);
  89         __common_field(int, pid);
  90         __common_field(int, lock_depth);
  91
  92         return ret;
  93 }
  94 EXPORT_SYMBOL_GPL(trace_define_common_fields);
  95
  96 void trace_destroy_fields(struct ftrace_event_call *call)
  97 {
  98         struct ftrace_event_field *field, *next;
  99
 100         list_for_each_entry_safe(field, next, &call->fields, link) {
 101                 list_del(&field->link);
 102                 kfree(field->type);
 103                 kfree(field->name);
 104                 kfree(field);
 105         }
 106 }
 107
 108 int trace_event_raw_init(struct ftrace_event_call *call)
 109 {
 110         int id;
 111
 112         id = register_ftrace_event(call->event);
 113         if (!id)
 114                 return -ENODEV;
 115         call->id = id;
 116         INIT_LIST_HEAD(&call->fields);
 117
 118         return 0;
 119 }
 120 EXPORT_SYMBOL_GPL(trace_event_raw_init);
 121
 122 static void ftrace_event_enable_disable(struct ftrace_event_call *call,
 123                                         int enable)
 124 {
 125         switch (enable) {
 126         case 0:
 127                 if (call->enabled) {
 128                         call->enabled = 0;
 129                         tracing_stop_cmdline_record();
 130                         call->unregfunc(call);
 131                 }
 132                 break;
 133         case 1:
 134                 if (!call->enabled) {
 135                         call->enabled = 1;
 136                         tracing_start_cmdline_record();
 137                         call->regfunc(call);
 138                 }
 139                 break;
 140         }
 141 }
 142
 143 static void ftrace_clear_events(void)
 144 {
 145         struct ftrace_event_call *call;
 146
 147         mutex_lock(&event_mutex);
 148         list_for_each_entry(call, &ftrace_events, list) {
 149                 ftrace_event_enable_disable(call, 0);
 150         }
 151         mutex_unlock(&event_mutex);
 152 }
 153
 154 /*
 155  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 156  */
 157 static int __ftrace_set_clr_event(const char *match, const char *sub,
 158                                   const char *event, int set)
 159 {
 160         struct ftrace_event_call *call;
 161         int ret = -EINVAL;
 162
 163         mutex_lock(&event_mutex);
 164         list_for_each_entry(call, &ftrace_events, list) {
 165
 166                 if (!call->name || !call->regfunc)
 167                         continue;
 168
 169                 if (match &&
 170                     strcmp(match, call->name) != 0 &&
 171                     strcmp(match, call->system) != 0)
 172                         continue;
 173
 174                 if (sub && strcmp(sub, call->system) != 0)
 175                         continue;
 176
 177                 if (event && strcmp(event, call->name) != 0)
 178                         continue;
 179
 180                 ftrace_event_enable_disable(call, set);
 181
 182                 ret = 0;
 183         }
 184         mutex_unlock(&event_mutex);
 185
 186         return ret;
 187 }
 188
 189 static int ftrace_set_clr_event(char *buf, int set)
 190 {
 191         char *event = NULL, *sub = NULL, *match;
 192
 193         /*
 194          * The buf format can be <subsystem>:<event-name>
 195          *  *:<event-name> means any event by that name.
 196          *  :<event-name> is the same.
 197          *
 198          *  <subsystem>:* means all events in that subsystem
 199          *  <subsystem>: means the same.
 200          *
 201          *  <name> (no ':') means all events in a subsystem with
 202          *  the name <name> or any event that matches <name>
 203          */
 204
 205         match = strsep(&buf, ":");
 206         if (buf) {
 207                 sub = match;
 208                 event = buf;
 209                 match = NULL;
 210
 211                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 212                         sub = NULL;
 213                 if (!strlen(event) || strcmp(event, "*") == 0)
 214                         event = NULL;
 215         }
 216
 217         return __ftrace_set_clr_event(match, sub, event, set);
 218 }
 219
 220 /**
 221  * trace_set_clr_event - enable or disable an event
 222  * @system: system name to match (NULL for any system)
 223  * @event: event name to match (NULL for all events, within system)
 224  * @set: 1 to enable, 0 to disable
 225  *
 226  * This is a way for other parts of the kernel to enable or disable
 227  * event recording.
 228  *
 229  * Returns 0 on success, -EINVAL if the parameters do not match any
 230  * registered events.
 231  */
 232 int trace_set_clr_event(const char *system, const char *event, int set)
 233 {
 234         return __ftrace_set_clr_event(NULL, system, event, set);
 235 }
 236
 237 /* 128 should be much more than enough */
 238 #define EVENT_BUF_SIZE          127
 239
 240 static ssize_t
 241 ftrace_event_write(struct file *file, const char __user *ubuf,
 242                    size_t cnt, loff_t *ppos)
 243 {
 244         struct trace_parser parser;
 245         ssize_t read, ret;
 246
 247         if (!cnt)
 248                 return 0;
 249
 250         ret = tracing_update_buffers();
 251         if (ret < 0)
 252                 return ret;
 253
 254         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 255                 return -ENOMEM;
 256
 257         read = trace_get_user(&parser, ubuf, cnt, ppos);
 258
 259         if (read >= 0 && trace_parser_loaded((&parser))) {
 260                 int set = 1;
 261
 262                 if (*parser.buffer == '!')
 263                         set = 0;
 264
 265                 parser.buffer[parser.idx] = 0;
 266
 267                 ret = ftrace_set_clr_event(parser.buffer + !set, set);
 268                 if (ret)
 269                         goto out_put;
 270         }
 271
 272         ret = read;
 273
 274  out_put:
 275         trace_parser_put(&parser);
 276
 277         return ret;
 278 }
 279
 280 static void *
 281 t_next(struct seq_file *m, void *v, loff_t *pos)
 282 {
 283         struct ftrace_event_call *call = v;
 284
 285         (*pos)++;
 286
 287         list_for_each_entry_continue(call, &ftrace_events, list) {
 288                 /*
 289                  * The ftrace subsystem is for showing formats only.
 290                  * They can not be enabled or disabled via the event files.
 291                  */
 292                 if (call->regfunc)
 293                         return call;
 294         }
 295
 296         return NULL;
 297 }
 298
 299 static void *t_start(struct seq_file *m, loff_t *pos)
 300 {
 301         struct ftrace_event_call *call;
 302         loff_t l;
 303
 304         mutex_lock(&event_mutex);
 305
 306         call = list_entry(&ftrace_events, struct ftrace_event_call, list);
 307         for (l = 0; l <= *pos; ) {
 308                 call = t_next(m, call, &l);
 309                 if (!call)
 310                         break;
 311         }
 312         return call;
 313 }
 314
 315 static void *
 316 s_next(struct seq_file *m, void *v, loff_t *pos)
 317 {
 318         struct ftrace_event_call *call = v;
 319
 320         (*pos)++;
 321
 322         list_for_each_entry_continue(call, &ftrace_events, list) {
 323                 if (call->enabled)
 324                         return call;
 325         }
 326
 327         return NULL;
 328 }
 329
 330 static void *s_start(struct seq_file *m, loff_t *pos)
 331 {
 332         struct ftrace_event_call *call;
 333         loff_t l;
 334
 335         mutex_lock(&event_mutex);
 336
 337         call = list_entry(&ftrace_events, struct ftrace_event_call, list);
 338         for (l = 0; l <= *pos; ) {
 339                 call = s_next(m, call, &l);
 340                 if (!call)
 341                         break;
 342         }
 343         return call;
 344 }
 345
 346 static int t_show(struct seq_file *m, void *v)
 347 {
 348         struct ftrace_event_call *call = v;
 349
 350         if (strcmp(call->system, TRACE_SYSTEM) != 0)
 351                 seq_printf(m, "%s:", call->system);
 352         seq_printf(m, "%s\n", call->name);
 353
 354         return 0;
 355 }
 356
 357 static void t_stop(struct seq_file *m, void *p)
 358 {
 359         mutex_unlock(&event_mutex);
 360 }
 361
 362 static int
 363 ftrace_event_seq_open(struct inode *inode, struct file *file)
 364 {
 365         const struct seq_operations *seq_ops;
 366
 367         if ((file->f_mode & FMODE_WRITE) &&
 368             (file->f_flags & O_TRUNC))
 369                 ftrace_clear_events();
 370
 371         seq_ops = inode->i_private;
 372         return seq_open(file, seq_ops);
 373 }
 374
 375 static ssize_t
 376 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 377                   loff_t *ppos)
 378 {
 379         struct ftrace_event_call *call = filp->private_data;
 380         char *buf;
 381
 382         if (call->enabled)
 383                 buf = "1\n";
 384         else
 385                 buf = "0\n";
 386
 387         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 388 }
 389
 390 static ssize_t
 391 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 392                    loff_t *ppos)
 393 {
 394         struct ftrace_event_call *call = filp->private_data;
 395         char buf[64];
 396         unsigned long val;
 397         int ret;
 398
 399         if (cnt >= sizeof(buf))
 400                 return -EINVAL;
 401
 402         if (copy_from_user(&buf, ubuf, cnt))
 403                 return -EFAULT;
 404
 405         buf[cnt] = 0;
 406
 407         ret = strict_strtoul(buf, 10, &val);
 408         if (ret < 0)
 409                 return ret;
 410
 411         ret = tracing_update_buffers();
 412         if (ret < 0)
 413                 return ret;
 414
 415         switch (val) {
 416         case 0:
 417         case 1:
 418                 mutex_lock(&event_mutex);
 419                 ftrace_event_enable_disable(call, val);
 420                 mutex_unlock(&event_mutex);
 421                 break;
 422
 423         default:
 424                 return -EINVAL;
 425         }
 426
 427         *ppos += cnt;
 428
 429         return cnt;
 430 }
 431
 432 static ssize_t
 433 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 434                    loff_t *ppos)
 435 {
 436         const char set_to_char[4] = { '?', '0', '1', 'X' };
 437         const char *system = filp->private_data;
 438         struct ftrace_event_call *call;
 439         char buf[2];
 440         int set = 0;
 441         int ret;
 442
 443         mutex_lock(&event_mutex);
 444         list_for_each_entry(call, &ftrace_events, list) {
 445                 if (!call->name || !call->regfunc)
 446                         continue;
 447
 448                 if (system && strcmp(call->system, system) != 0)
 449                         continue;
 450
 451                 /*
 452                  * We need to find out if all the events are set
 453                  * or if all events or cleared, or if we have
 454                  * a mixture.
 455                  */
 456                 set |= (1 << !!call->enabled);
 457
 458                 /*
 459                  * If we have a mixture, no need to look further.
 460                  */
 461                 if (set == 3)
 462                         break;
 463         }
 464         mutex_unlock(&event_mutex);
 465
 466         buf[0] = set_to_char[set];
 467         buf[1] = '\n';
 468
 469         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 470
 471         return ret;
 472 }
 473
 474 static ssize_t
 475 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 476                     loff_t *ppos)
 477 {
 478         const char *system = filp->private_data;
 479         unsigned long val;
 480         char buf[64];
 481         ssize_t ret;
 482
 483         if (cnt >= sizeof(buf))
 484                 return -EINVAL;
 485
 486         if (copy_from_user(&buf, ubuf, cnt))
 487                 return -EFAULT;
 488
 489         buf[cnt] = 0;
 490
 491         ret = strict_strtoul(buf, 10, &val);
 492         if (ret < 0)
 493                 return ret;
 494
 495         ret = tracing_update_buffers();
 496         if (ret < 0)
 497                 return ret;
 498
 499         if (val != 0 && val != 1)
 500                 return -EINVAL;
 501
 502         ret = __ftrace_set_clr_event(NULL, system, NULL, val);
 503         if (ret)
 504                 goto out;
 505
 506         ret = cnt;
 507
 508 out:
 509         *ppos += cnt;
 510
 511         return ret;
 512 }
 513
 514 extern char *__bad_type_size(void);
 515
 516 #undef FIELD
 517 #define FIELD(type, name)                                               \
 518         sizeof(type) != sizeof(field.name) ? __bad_type_size() :        \
 519         #type, "common_" #name, offsetof(typeof(field), name),          \
 520                 sizeof(field.name), is_signed_type(type)
 521
 522 static int trace_write_header(struct trace_seq *s)
 523 {
 524         struct trace_entry field;
 525
 526         /* struct trace_entry */
 527         return trace_seq_printf(s,
 528                         "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\tsigned:%u;\n"
 529                         "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\tsigned:%u;\n"
 530                         "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\tsigned:%u;\n"
 531                         "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\tsigned:%u;\n"
 532                         "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\tsigned:%u;\n"
 533                         "\n",
 534                         FIELD(unsigned short, type),
 535                         FIELD(unsigned char, flags),
 536                         FIELD(unsigned char, preempt_count),
 537                         FIELD(int, pid),
 538                         FIELD(int, lock_depth));
 539 }
 540
 541 static ssize_t
 542 event_format_read(struct file *filp, char __user *ubuf, size_t cnt,
 543                   loff_t *ppos)
 544 {
 545         struct ftrace_event_call *call = filp->private_data;
 546         struct trace_seq *s;
 547         char *buf;
 548         int r;
 549
 550         if (*ppos)
 551                 return 0;
 552
 553         s = kmalloc(sizeof(*s), GFP_KERNEL);
 554         if (!s)
 555                 return -ENOMEM;
 556
 557         trace_seq_init(s);
 558
 559         /* If any of the first writes fail, so will the show_format. */
 560
 561         trace_seq_printf(s, "name: %s\n", call->name);
 562         trace_seq_printf(s, "ID: %d\n", call->id);
 563         trace_seq_printf(s, "format:\n");
 564         trace_write_header(s);
 565
 566         r = call->show_format(call, s);
 567         if (!r) {
 568                 /*
 569                  * ug!  The format output is bigger than a PAGE!!
 570                  */
 571                 buf = "FORMAT TOO BIG\n";
 572                 r = simple_read_from_buffer(ubuf, cnt, ppos,
 573                                               buf, strlen(buf));
 574                 goto out;
 575         }
 576
 577         r = simple_read_from_buffer(ubuf, cnt, ppos,
 578                                     s->buffer, s->len);
 579  out:
 580         kfree(s);
 581         return r;
 582 }
 583
 584 static ssize_t
 585 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 586 {
 587         struct ftrace_event_call *call = filp->private_data;
 588         struct trace_seq *s;
 589         int r;
 590
 591         if (*ppos)
 592                 return 0;
 593
 594         s = kmalloc(sizeof(*s), GFP_KERNEL);
 595         if (!s)
 596                 return -ENOMEM;
 597
 598         trace_seq_init(s);
 599         trace_seq_printf(s, "%d\n", call->id);
 600
 601         r = simple_read_from_buffer(ubuf, cnt, ppos,
 602                                     s->buffer, s->len);
 603         kfree(s);
 604         return r;
 605 }
 606
 607 static ssize_t
 608 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 609                   loff_t *ppos)
 610 {
 611         struct ftrace_event_call *call = filp->private_data;
 612         struct trace_seq *s;
 613         int r;
 614
 615         if (*ppos)
 616                 return 0;
 617
 618         s = kmalloc(sizeof(*s), GFP_KERNEL);
 619         if (!s)
 620                 return -ENOMEM;
 621
 622         trace_seq_init(s);
 623
 624         print_event_filter(call, s);
 625         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 626
 627         kfree(s);
 628
 629         return r;
 630 }
 631
 632 static ssize_t
 633 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
 634                    loff_t *ppos)
 635 {
 636         struct ftrace_event_call *call = filp->private_data;
 637         char *buf;
 638         int err;
 639
 640         if (cnt >= PAGE_SIZE)
 641                 return -EINVAL;
 642
 643         buf = (char *)__get_free_page(GFP_TEMPORARY);
 644         if (!buf)
 645                 return -ENOMEM;
 646
 647         if (copy_from_user(buf, ubuf, cnt)) {
 648                 free_page((unsigned long) buf);
 649                 return -EFAULT;
 650         }
 651         buf[cnt] = '\0';
 652
 653         err = apply_event_filter(call, buf);
 654         free_page((unsigned long) buf);
 655         if (err < 0)
 656                 return err;
 657
 658         *ppos += cnt;
 659
 660         return cnt;
 661 }
 662
 663 static ssize_t
 664 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 665                       loff_t *ppos)
 666 {
 667         struct event_subsystem *system = filp->private_data;
 668         struct trace_seq *s;
 669         int r;
 670
 671         if (*ppos)
 672                 return 0;
 673
 674         s = kmalloc(sizeof(*s), GFP_KERNEL);
 675         if (!s)
 676                 return -ENOMEM;
 677
 678         trace_seq_init(s);
 679
 680         print_subsystem_event_filter(system, s);
 681         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 682
 683         kfree(s);
 684
 685         return r;
 686 }
 687
 688 static ssize_t
 689 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
 690                        loff_t *ppos)
 691 {
 692         struct event_subsystem *system = filp->private_data;
 693         char *buf;
 694         int err;
 695
 696         if (cnt >= PAGE_SIZE)
 697                 return -EINVAL;
 698
 699         buf = (char *)__get_free_page(GFP_TEMPORARY);
 700         if (!buf)
 701                 return -ENOMEM;
 702
 703         if (copy_from_user(buf, ubuf, cnt)) {
 704                 free_page((unsigned long) buf);
 705                 return -EFAULT;
 706         }
 707         buf[cnt] = '\0';
 708
 709         err = apply_subsystem_event_filter(system, buf);
 710         free_page((unsigned long) buf);
 711         if (err < 0)
 712                 return err;
 713
 714         *ppos += cnt;
 715
 716         return cnt;
 717 }
 718
 719 static ssize_t
 720 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 721 {
 722         int (*func)(struct trace_seq *s) = filp->private_data;
 723         struct trace_seq *s;
 724         int r;
 725
 726         if (*ppos)
 727                 return 0;
 728
 729         s = kmalloc(sizeof(*s), GFP_KERNEL);
 730         if (!s)
 731                 return -ENOMEM;
 732
 733         trace_seq_init(s);
 734
 735         func(s);
 736         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 737
 738         kfree(s);
 739
 740         return r;
 741 }
 742
 743 static const struct seq_operations show_event_seq_ops = {
 744         .start = t_start,
 745         .next = t_next,
 746         .show = t_show,
 747         .stop = t_stop,
 748 };
 749
 750 static const struct seq_operations show_set_event_seq_ops = {
 751         .start = s_start,
 752         .next = s_next,
 753         .show = t_show,
 754         .stop = t_stop,
 755 };
 756
 757 static const struct file_operations ftrace_avail_fops = {
 758         .open = ftrace_event_seq_open,
 759         .read = seq_read,
 760         .llseek = seq_lseek,
 761         .release = seq_release,
 762 };
 763
 764 static const struct file_operations ftrace_set_event_fops = {
 765         .open = ftrace_event_seq_open,
 766         .read = seq_read,
 767         .write = ftrace_event_write,
 768         .llseek = seq_lseek,
 769         .release = seq_release,
 770 };
 771
 772 static const struct file_operations ftrace_enable_fops = {
 773         .open = tracing_open_generic,
 774         .read = event_enable_read,
 775         .write = event_enable_write,
 776 };
 777
 778 static const struct file_operations ftrace_event_format_fops = {
 779         .open = tracing_open_generic,
 780         .read = event_format_read,
 781 };
 782
 783 static const struct file_operations ftrace_event_id_fops = {
 784         .open = tracing_open_generic,
 785         .read = event_id_read,
 786 };
 787
 788 static const struct file_operations ftrace_event_filter_fops = {
 789         .open = tracing_open_generic,
 790         .read = event_filter_read,
 791         .write = event_filter_write,
 792 };
 793
 794 static const struct file_operations ftrace_subsystem_filter_fops = {
 795         .open = tracing_open_generic,
 796         .read = subsystem_filter_read,
 797         .write = subsystem_filter_write,
 798 };
 799
 800 static const struct file_operations ftrace_system_enable_fops = {
 801         .open = tracing_open_generic,
 802         .read = system_enable_read,
 803         .write = system_enable_write,
 804 };
 805
 806 static const struct file_operations ftrace_show_header_fops = {
 807         .open = tracing_open_generic,
 808         .read = show_header,
 809 };
 810
 811 static struct dentry *event_trace_events_dir(void)
 812 {
 813         static struct dentry *d_tracer;
 814         static struct dentry *d_events;
 815
 816         if (d_events)
 817                 return d_events;
 818
 819         d_tracer = tracing_init_dentry();
 820         if (!d_tracer)
 821                 return NULL;
 822
 823         d_events = debugfs_create_dir("events", d_tracer);
 824         if (!d_events)
 825                 pr_warning("Could not create debugfs "
 826                            "'events' directory\n");
 827
 828         return d_events;
 829 }
 830
 831 static LIST_HEAD(event_subsystems);
 832
 833 static struct dentry *
 834 event_subsystem_dir(const char *name, struct dentry *d_events)
 835 {
 836         struct event_subsystem *system;
 837         struct dentry *entry;
 838
 839         /* First see if we did not already create this dir */
 840         list_for_each_entry(system, &event_subsystems, list) {
 841                 if (strcmp(system->name, name) == 0) {
 842                         system->nr_events++;
 843                         return system->entry;
 844                 }
 845         }
 846
 847         /* need to create new entry */
 848         system = kmalloc(sizeof(*system), GFP_KERNEL);
 849         if (!system) {
 850                 pr_warning("No memory to create event subsystem %s\n",
 851                            name);
 852                 return d_events;
 853         }
 854
 855         system->entry = debugfs_create_dir(name, d_events);
 856         if (!system->entry) {
 857                 pr_warning("Could not create event subsystem %s\n",
 858                            name);
 859                 kfree(system);
 860                 return d_events;
 861         }
 862
 863         system->nr_events = 1;
 864         system->name = kstrdup(name, GFP_KERNEL);
 865         if (!system->name) {
 866                 debugfs_remove(system->entry);
 867                 kfree(system);
 868                 return d_events;
 869         }
 870
 871         list_add(&system->list, &event_subsystems);
 872
 873         system->filter = NULL;
 874
 875         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
 876         if (!system->filter) {
 877                 pr_warning("Could not allocate filter for subsystem "
 878                            "'%s'\n", name);
 879                 return system->entry;
 880         }
 881
 882         entry = debugfs_create_file("filter", 0644, system->entry, system,
 883                                     &ftrace_subsystem_filter_fops);
 884         if (!entry) {
 885                 kfree(system->filter);
 886                 system->filter = NULL;
 887                 pr_warning("Could not create debugfs "
 888                            "'%s/filter' entry\n", name);
 889         }
 890
 891         trace_create_file("enable", 0644, system->entry,
 892                           (void *)system->name,
 893                           &ftrace_system_enable_fops);
 894
 895         return system->entry;
 896 }
 897
 898 static int
 899 event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
 900                  const struct file_operations *id,
 901                  const struct file_operations *enable,
 902                  const struct file_operations *filter,
 903                  const struct file_operations *format)
 904 {
 905         int ret;
 906
 907         /*
 908          * If the trace point header did not define TRACE_SYSTEM
 909          * then the system would be called "TRACE_SYSTEM".
 910          */
 911         if (strcmp(call->system, TRACE_SYSTEM) != 0)
 912                 d_events = event_subsystem_dir(call->system, d_events);
 913
 914         call->dir = debugfs_create_dir(call->name, d_events);
 915         if (!call->dir) {
 916                 pr_warning("Could not create debugfs "
 917                            "'%s' directory\n", call->name);
 918                 return -1;
 919         }
 920
 921         if (call->regfunc)
 922                 trace_create_file("enable", 0644, call->dir, call,
 923                                   enable);
 924
 925         if (call->id && call->profile_enable)
 926                 trace_create_file("id", 0444, call->dir, call,
 927                                   id);
 928
 929         if (call->define_fields) {
 930                 ret = call->define_fields(call);
 931                 if (ret < 0) {
 932                         pr_warning("Could not initialize trace point"
 933                                    " events/%s\n", call->name);
 934                         return ret;
 935                 }
 936                 trace_create_file("filter", 0644, call->dir, call,
 937                                   filter);
 938         }
 939
 940         /* A trace may not want to export its format */
 941         if (!call->show_format)
 942                 return 0;
 943
 944         trace_create_file("format", 0444, call->dir, call,
 945                           format);
 946
 947         return 0;
 948 }
 949
 950 static int __trace_add_event_call(struct ftrace_event_call *call)
 951 {
 952         struct dentry *d_events;
 953         int ret;
 954
 955         if (!call->name)
 956                 return -EINVAL;
 957
 958         if (call->raw_init) {
 959                 ret = call->raw_init(call);
 960                 if (ret < 0) {
 961                         if (ret != -ENOSYS)
 962                                 pr_warning("Could not initialize trace "
 963                                 "events/%s\n", call->name);
 964                         return ret;
 965                 }
 966         }
 967
 968         d_events = event_trace_events_dir();
 969         if (!d_events)
 970                 return -ENOENT;
 971
 972         ret = event_create_dir(call, d_events, &ftrace_event_id_fops,
 973                                 &ftrace_enable_fops, &ftrace_event_filter_fops,
 974                                 &ftrace_event_format_fops);
 975         if (!ret)
 976                 list_add(&call->list, &ftrace_events);
 977
 978         return ret;
 979 }
 980
 981 /* Add an additional event_call dynamically */
 982 int trace_add_event_call(struct ftrace_event_call *call)
 983 {
 984         int ret;
 985         mutex_lock(&event_mutex);
 986         ret = __trace_add_event_call(call);
 987         mutex_unlock(&event_mutex);
 988         return ret;
 989 }
 990
 991 static void remove_subsystem_dir(const char *name)
 992 {
 993         struct event_subsystem *system;
 994
 995         if (strcmp(name, TRACE_SYSTEM) == 0)
 996                 return;
 997
 998         list_for_each_entry(system, &event_subsystems, list) {
 999                 if (strcmp(system->name, name) == 0) {
1000                         if (!--system->nr_events) {
1001                                 struct event_filter *filter = system->filter;
1002
1003                                 debugfs_remove_recursive(system->entry);
1004                                 list_del(&system->list);
1005                                 if (filter) {
1006                                         kfree(filter->filter_string);
1007                                         kfree(filter);
1008                                 }
1009                                 kfree(system->name);
1010                                 kfree(system);
1011                         }
1012                         break;
1013                 }
1014         }
1015 }
1016
1017 /*
1018  * Must be called under locking both of event_mutex and trace_event_mutex.
1019  */
1020 static void __trace_remove_event_call(struct ftrace_event_call *call)
1021 {
1022         ftrace_event_enable_disable(call, 0);
1023         if (call->event)
1024                 __unregister_ftrace_event(call->event);
1025         debugfs_remove_recursive(call->dir);
1026         list_del(&call->list);
1027         trace_destroy_fields(call);
1028         destroy_preds(call);
1029         remove_subsystem_dir(call->system);
1030 }
1031
1032 /* Remove an event_call */
1033 void trace_remove_event_call(struct ftrace_event_call *call)
1034 {
1035         mutex_lock(&event_mutex);
1036         down_write(&trace_event_mutex);
1037         __trace_remove_event_call(call);
1038         up_write(&trace_event_mutex);
1039         mutex_unlock(&event_mutex);
1040 }
1041
1042 #define for_each_event(event, start, end)                       \
1043         for (event = start;                                     \
1044              (unsigned long)event < (unsigned long)end;         \
1045              event++)
1046
1047 #ifdef CONFIG_MODULES
1048
1049 static LIST_HEAD(ftrace_module_file_list);
1050
1051 /*
1052  * Modules must own their file_operations to keep up with
1053  * reference counting.
1054  */
1055 struct ftrace_module_file_ops {
1056         struct list_head                list;
1057         struct module                   *mod;
1058         struct file_operations          id;
1059         struct file_operations          enable;
1060         struct file_operations          format;
1061         struct file_operations          filter;
1062 };
1063
1064 static struct ftrace_module_file_ops *
1065 trace_create_file_ops(struct module *mod)
1066 {
1067         struct ftrace_module_file_ops *file_ops;
1068
1069         /*
1070          * This is a bit of a PITA. To allow for correct reference
1071          * counting, modules must "own" their file_operations.
1072          * To do this, we allocate the file operations that will be
1073          * used in the event directory.
1074          */
1075
1076         file_ops = kmalloc(sizeof(*file_ops), GFP_KERNEL);
1077         if (!file_ops)
1078                 return NULL;
1079
1080         file_ops->mod = mod;
1081
1082         file_ops->id = ftrace_event_id_fops;
1083         file_ops->id.owner = mod;
1084
1085         file_ops->enable = ftrace_enable_fops;
1086         file_ops->enable.owner = mod;
1087
1088         file_ops->filter = ftrace_event_filter_fops;
1089         file_ops->filter.owner = mod;
1090
1091         file_ops->format = ftrace_event_format_fops;
1092         file_ops->format.owner = mod;
1093
1094         list_add(&file_ops->list, &ftrace_module_file_list);
1095
1096         return file_ops;
1097 }
1098
1099 static void trace_module_add_events(struct module *mod)
1100 {
1101         struct ftrace_module_file_ops *file_ops = NULL;
1102         struct ftrace_event_call *call, *start, *end;
1103         struct dentry *d_events;
1104         int ret;
1105
1106         start = mod->trace_events;
1107         end = mod->trace_events + mod->num_trace_events;
1108
1109         if (start == end)
1110                 return;
1111
1112         d_events = event_trace_events_dir();
1113         if (!d_events)
1114                 return;
1115
1116         for_each_event(call, start, end) {
1117                 /* The linker may leave blanks */
1118                 if (!call->name)
1119                         continue;
1120                 if (call->raw_init) {
1121                         ret = call->raw_init(call);
1122                         if (ret < 0) {
1123                                 if (ret != -ENOSYS)
1124                                         pr_warning("Could not initialize trace "
1125                                         "point events/%s\n", call->name);
1126                                 continue;
1127                         }
1128                 }
1129                 /*
1130                  * This module has events, create file ops for this module
1131                  * if not already done.
1132                  */
1133                 if (!file_ops) {
1134                         file_ops = trace_create_file_ops(mod);
1135                         if (!file_ops)
1136                                 return;
1137                 }
1138                 call->mod = mod;
1139                 ret = event_create_dir(call, d_events,
1140                                        &file_ops->id, &file_ops->enable,
1141                                        &file_ops->filter, &file_ops->format);
1142                 if (!ret)
1143                         list_add(&call->list, &ftrace_events);
1144         }
1145 }
1146
1147 static void trace_module_remove_events(struct module *mod)
1148 {
1149         struct ftrace_module_file_ops *file_ops;
1150         struct ftrace_event_call *call, *p;
1151         bool found = false;
1152
1153         down_write(&trace_event_mutex);
1154         list_for_each_entry_safe(call, p, &ftrace_events, list) {
1155                 if (call->mod == mod) {
1156                         found = true;
1157                         __trace_remove_event_call(call);
1158                 }
1159         }
1160
1161         /* Now free the file_operations */
1162         list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1163                 if (file_ops->mod == mod)
1164                         break;
1165         }
1166         if (&file_ops->list != &ftrace_module_file_list) {
1167                 list_del(&file_ops->list);
1168                 kfree(file_ops);
1169         }
1170
1171         /*
1172          * It is safest to reset the ring buffer if the module being unloaded
1173          * registered any events.
1174          */
1175         if (found)
1176                 tracing_reset_current_online_cpus();
1177         up_write(&trace_event_mutex);
1178 }
1179
1180 static int trace_module_notify(struct notifier_block *self,
1181                                unsigned long val, void *data)
1182 {
1183         struct module *mod = data;
1184
1185         mutex_lock(&event_mutex);
1186         switch (val) {
1187         case MODULE_STATE_COMING:
1188                 trace_module_add_events(mod);
1189                 break;
1190         case MODULE_STATE_GOING:
1191                 trace_module_remove_events(mod);
1192                 break;
1193         }
1194         mutex_unlock(&event_mutex);
1195
1196         return 0;
1197 }
1198 #else
1199 static int trace_module_notify(struct notifier_block *self,
1200                                unsigned long val, void *data)
1201 {
1202         return 0;
1203 }
1204 #endif /* CONFIG_MODULES */
1205
1206 static struct notifier_block trace_module_nb = {
1207         .notifier_call = trace_module_notify,
1208         .priority = 0,
1209 };
1210
1211 extern struct ftrace_event_call __start_ftrace_events[];
1212 extern struct ftrace_event_call __stop_ftrace_events[];
1213
1214 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
1215
1216 static __init int setup_trace_event(char *str)
1217 {
1218         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
1219         ring_buffer_expanded = 1;
1220         tracing_selftest_disabled = 1;
1221
1222         return 1;
1223 }
1224 __setup("trace_event=", setup_trace_event);
1225
1226 static __init int event_trace_init(void)
1227 {
1228         struct ftrace_event_call *call;
1229         struct dentry *d_tracer;
1230         struct dentry *entry;
1231         struct dentry *d_events;
1232         int ret;
1233         char *buf = bootup_event_buf;
1234         char *token;
1235
1236         d_tracer = tracing_init_dentry();
1237         if (!d_tracer)
1238                 return 0;
1239
1240         entry = debugfs_create_file("available_events", 0444, d_tracer,
1241                                     (void *)&show_event_seq_ops,
1242                                     &ftrace_avail_fops);
1243         if (!entry)
1244                 pr_warning("Could not create debugfs "
1245                            "'available_events' entry\n");
1246
1247         entry = debugfs_create_file("set_event", 0644, d_tracer,
1248                                     (void *)&show_set_event_seq_ops,
1249                                     &ftrace_set_event_fops);
1250         if (!entry)
1251                 pr_warning("Could not create debugfs "
1252                            "'set_event' entry\n");
1253
1254         d_events = event_trace_events_dir();
1255         if (!d_events)
1256                 return 0;
1257
1258         /* ring buffer internal formats */
1259         trace_create_file("header_page", 0444, d_events,
1260                           ring_buffer_print_page_header,
1261                           &ftrace_show_header_fops);
1262
1263         trace_create_file("header_event", 0444, d_events,
1264                           ring_buffer_print_entry_header,
1265                           &ftrace_show_header_fops);
1266
1267         trace_create_file("enable", 0644, d_events,
1268                           NULL, &ftrace_system_enable_fops);
1269
1270         for_each_event(call, __start_ftrace_events, __stop_ftrace_events) {
1271                 /* The linker may leave blanks */
1272                 if (!call->name)
1273                         continue;
1274                 if (call->raw_init) {
1275                         ret = call->raw_init(call);
1276                         if (ret < 0) {
1277                                 if (ret != -ENOSYS)
1278                                         pr_warning("Could not initialize trace "
1279                                         "point events/%s\n", call->name);
1280                                 continue;
1281                         }
1282                 }
1283                 ret = event_create_dir(call, d_events, &ftrace_event_id_fops,
1284                                        &ftrace_enable_fops,
1285                                        &ftrace_event_filter_fops,
1286                                        &ftrace_event_format_fops);
1287                 if (!ret)
1288                         list_add(&call->list, &ftrace_events);
1289         }
1290
1291         while (true) {
1292                 token = strsep(&buf, ",");
1293
1294                 if (!token)
1295                         break;
1296                 if (!*token)
1297                         continue;
1298
1299                 ret = ftrace_set_clr_event(token, 1);
1300                 if (ret)
1301                         pr_warning("Failed to enable trace event: %s\n", token);
1302         }
1303
1304         ret = register_module_notifier(&trace_module_nb);
1305         if (ret)
1306                 pr_warning("Failed to register trace events module notifier\n");
1307
1308         return 0;
1309 }
1310 fs_initcall(event_trace_init);
1311
1312 #ifdef CONFIG_FTRACE_STARTUP_TEST
1313
1314 static DEFINE_SPINLOCK(test_spinlock);
1315 static DEFINE_SPINLOCK(test_spinlock_irq);
1316 static DEFINE_MUTEX(test_mutex);
1317
1318 static __init void test_work(struct work_struct *dummy)
1319 {
1320         spin_lock(&test_spinlock);
1321         spin_lock_irq(&test_spinlock_irq);
1322         udelay(1);
1323         spin_unlock_irq(&test_spinlock_irq);
1324         spin_unlock(&test_spinlock);
1325
1326         mutex_lock(&test_mutex);
1327         msleep(1);
1328         mutex_unlock(&test_mutex);
1329 }
1330
1331 static __init int event_test_thread(void *unused)
1332 {
1333         void *test_malloc;
1334
1335         test_malloc = kmalloc(1234, GFP_KERNEL);
1336         if (!test_malloc)
1337                 pr_info("failed to kmalloc\n");
1338
1339         schedule_on_each_cpu(test_work);
1340
1341         kfree(test_malloc);
1342
1343         set_current_state(TASK_INTERRUPTIBLE);
1344         while (!kthread_should_stop())
1345                 schedule();
1346
1347         return 0;
1348 }
1349
1350 /*
1351  * Do various things that may trigger events.
1352  */
1353 static __init void event_test_stuff(void)
1354 {
1355         struct task_struct *test_thread;
1356
1357         test_thread = kthread_run(event_test_thread, NULL, "test-events");
1358         msleep(1);
1359         kthread_stop(test_thread);
1360 }
1361
1362 /*
1363  * For every trace event defined, we will test each trace point separately,
1364  * and then by groups, and finally all trace points.
1365  */
1366 static __init void event_trace_self_tests(void)
1367 {
1368         struct ftrace_event_call *call;
1369         struct event_subsystem *system;
1370         int ret;
1371
1372         pr_info("Running tests on trace events:\n");
1373
1374         list_for_each_entry(call, &ftrace_events, list) {
1375
1376                 /* Only test those that have a regfunc */
1377                 if (!call->regfunc)
1378                         continue;
1379
1380 /*
1381  * Testing syscall events here is pretty useless, but
1382  * we still do it if configured. But this is time consuming.
1383  * What we really need is a user thread to perform the
1384  * syscalls as we test.
1385  */
1386 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
1387                 if (call->system &&
1388                     strcmp(call->system, "syscalls") == 0)
1389                         continue;
1390 #endif
1391
1392                 pr_info("Testing event %s: ", call->name);
1393
1394                 /*
1395                  * If an event is already enabled, someone is using
1396                  * it and the self test should not be on.
1397                  */
1398                 if (call->enabled) {
1399                         pr_warning("Enabled event during self test!\n");
1400                         WARN_ON_ONCE(1);
1401                         continue;
1402                 }
1403
1404                 ftrace_event_enable_disable(call, 1);
1405                 event_test_stuff();
1406                 ftrace_event_enable_disable(call, 0);
1407
1408                 pr_cont("OK\n");
1409         }
1410
1411         /* Now test at the sub system level */
1412
1413         pr_info("Running tests on trace event systems:\n");
1414
1415         list_for_each_entry(system, &event_subsystems, list) {
1416
1417                 /* the ftrace system is special, skip it */
1418                 if (strcmp(system->name, "ftrace") == 0)
1419                         continue;
1420
1421                 pr_info("Testing event system %s: ", system->name);
1422
1423                 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 1);
1424                 if (WARN_ON_ONCE(ret)) {
1425                         pr_warning("error enabling system %s\n",
1426                                    system->name);
1427                         continue;
1428                 }
1429
1430                 event_test_stuff();
1431
1432                 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 0);
1433                 if (WARN_ON_ONCE(ret))
1434                         pr_warning("error disabling system %s\n",
1435                                    system->name);
1436
1437                 pr_cont("OK\n");
1438         }
1439
1440         /* Test with all events enabled */
1441
1442         pr_info("Running tests on all trace events:\n");
1443         pr_info("Testing all events: ");
1444
1445         ret = __ftrace_set_clr_event(NULL, NULL, NULL, 1);
1446         if (WARN_ON_ONCE(ret)) {
1447                 pr_warning("error enabling all events\n");
1448                 return;
1449         }
1450
1451         event_test_stuff();
1452
1453         /* reset sysname */
1454         ret = __ftrace_set_clr_event(NULL, NULL, NULL, 0);
1455         if (WARN_ON_ONCE(ret)) {
1456                 pr_warning("error disabling all events\n");
1457                 return;
1458         }
1459
1460         pr_cont("OK\n");
1461 }
1462
1463 #ifdef CONFIG_FUNCTION_TRACER
1464
1465 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
1466
1467 static void
1468 function_test_events_call(unsigned long ip, unsigned long parent_ip)
1469 {
1470         struct ring_buffer_event *event;
1471         struct ring_buffer *buffer;
1472         struct ftrace_entry *entry;
1473         unsigned long flags;
1474         long disabled;
1475         int resched;
1476         int cpu;
1477         int pc;
1478
1479         pc = preempt_count();
1480         resched = ftrace_preempt_disable();
1481         cpu = raw_smp_processor_id();
1482         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
1483
1484         if (disabled != 1)
1485                 goto out;
1486
1487         local_save_flags(flags);
1488
1489         event = trace_current_buffer_lock_reserve(&buffer,
1490                                                   TRACE_FN, sizeof(*entry),
1491                                                   flags, pc);
1492         if (!event)
1493                 goto out;
1494         entry   = ring_buffer_event_data(event);
1495         entry->ip                       = ip;
1496         entry->parent_ip                = parent_ip;
1497
1498         trace_nowake_buffer_unlock_commit(buffer, event, flags, pc);
1499
1500  out:
1501         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
1502         ftrace_preempt_enable(resched);
1503 }
1504
1505 static struct ftrace_ops trace_ops __initdata  =
1506 {
1507         .func = function_test_events_call,
1508 };
1509
1510 static __init void event_trace_self_test_with_function(void)
1511 {
1512         register_ftrace_function(&trace_ops);
1513         pr_info("Running tests again, along with the function tracer\n");
1514         event_trace_self_tests();
1515         unregister_ftrace_function(&trace_ops);
1516 }
1517 #else
1518 static __init void event_trace_self_test_with_function(void)
1519 {
1520 }
1521 #endif
1522
1523 static __init int event_trace_self_tests_init(void)
1524 {
1525         if (!tracing_selftest_disabled) {
1526                 event_trace_self_tests();
1527                 event_trace_self_test_with_function();
1528         }
1529
1530         return 0;
1531 }
1532
1533 late_initcall(event_trace_self_tests_init);
1534
1535 #endif