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