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