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