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