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         struct ftrace_event_call *call = NULL;
 304         loff_t l;
 305
 306         mutex_lock(&event_mutex);
 307
 308         m->private = ftrace_events.next;
 309         for (l = 0; l <= *pos; ) {
 310                 call = t_next(m, NULL, &l);
 311                 if (!call)
 312                         break;
 313         }
 314         return call;
 315 }
 316
 317 static void *
 318 s_next(struct seq_file *m, void *v, loff_t *pos)
 319 {
 320         struct list_head *list = m->private;
 321         struct ftrace_event_call *call;
 322
 323         (*pos)++;
 324
 325  retry:
 326         if (list == &ftrace_events)
 327                 return NULL;
 328
 329         call = list_entry(list, struct ftrace_event_call, list);
 330
 331         if (!call->enabled) {
 332                 list = list->next;
 333                 goto retry;
 334         }
 335
 336         m->private = list->next;
 337
 338         return call;
 339 }
 340
 341 static void *s_start(struct seq_file *m, loff_t *pos)
 342 {
 343         struct ftrace_event_call *call = NULL;
 344         loff_t l;
 345
 346         mutex_lock(&event_mutex);
 347
 348         m->private = ftrace_events.next;
 349         for (l = 0; l <= *pos; ) {
 350                 call = s_next(m, NULL, &l);
 351                 if (!call)
 352                         break;
 353         }
 354         return call;
 355 }
 356
 357 static int t_show(struct seq_file *m, void *v)
 358 {
 359         struct ftrace_event_call *call = v;
 360
 361         if (strcmp(call->system, TRACE_SYSTEM) != 0)
 362                 seq_printf(m, "%s:", call->system);
 363         seq_printf(m, "%s\n", call->name);
 364
 365         return 0;
 366 }
 367
 368 static void t_stop(struct seq_file *m, void *p)
 369 {
 370         mutex_unlock(&event_mutex);
 371 }
 372
 373 static int
 374 ftrace_event_seq_open(struct inode *inode, struct file *file)
 375 {
 376         const struct seq_operations *seq_ops;
 377
 378         if ((file->f_mode & FMODE_WRITE) &&
 379             !(file->f_flags & O_APPEND))
 380                 ftrace_clear_events();
 381
 382         seq_ops = inode->i_private;
 383         return seq_open(file, seq_ops);
 384 }
 385
 386 static ssize_t
 387 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 388                   loff_t *ppos)
 389 {
 390         struct ftrace_event_call *call = filp->private_data;
 391         char *buf;
 392
 393         if (call->enabled)
 394                 buf = "1\n";
 395         else
 396                 buf = "0\n";
 397
 398         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 399 }
 400
 401 static ssize_t
 402 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 403                    loff_t *ppos)
 404 {
 405         struct ftrace_event_call *call = filp->private_data;
 406         char buf[64];
 407         unsigned long val;
 408         int ret;
 409
 410         if (cnt >= sizeof(buf))
 411                 return -EINVAL;
 412
 413         if (copy_from_user(&buf, ubuf, cnt))
 414                 return -EFAULT;
 415
 416         buf[cnt] = 0;
 417
 418         ret = strict_strtoul(buf, 10, &val);
 419         if (ret < 0)
 420                 return ret;
 421
 422         ret = tracing_update_buffers();
 423         if (ret < 0)
 424                 return ret;
 425
 426         switch (val) {
 427         case 0:
 428         case 1:
 429                 mutex_lock(&event_mutex);
 430                 ftrace_event_enable_disable(call, val);
 431                 mutex_unlock(&event_mutex);
 432                 break;
 433
 434         default:
 435                 return -EINVAL;
 436         }
 437
 438         *ppos += cnt;
 439
 440         return cnt;
 441 }
 442
 443 static ssize_t
 444 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 445                    loff_t *ppos)
 446 {
 447         const char set_to_char[4] = { '?', '0', '1', 'X' };
 448         const char *system = filp->private_data;
 449         struct ftrace_event_call *call;
 450         char buf[2];
 451         int set = 0;
 452         int ret;
 453
 454         mutex_lock(&event_mutex);
 455         list_for_each_entry(call, &ftrace_events, list) {
 456                 if (!call->name || !call->regfunc)
 457                         continue;
 458
 459                 if (system && strcmp(call->system, system) != 0)
 460                         continue;
 461
 462                 /*
 463                  * We need to find out if all the events are set
 464                  * or if all events or cleared, or if we have
 465                  * a mixture.
 466                  */
 467                 set |= (1 << !!call->enabled);
 468
 469                 /*
 470                  * If we have a mixture, no need to look further.
 471                  */
 472                 if (set == 3)
 473                         break;
 474         }
 475         mutex_unlock(&event_mutex);
 476
 477         buf[0] = set_to_char[set];
 478         buf[1] = '\n';
 479
 480         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 481
 482         return ret;
 483 }
 484
 485 static ssize_t
 486 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 487                     loff_t *ppos)
 488 {
 489         const char *system = filp->private_data;
 490         unsigned long val;
 491         char buf[64];
 492         ssize_t ret;
 493
 494         if (cnt >= sizeof(buf))
 495                 return -EINVAL;
 496
 497         if (copy_from_user(&buf, ubuf, cnt))
 498                 return -EFAULT;
 499
 500         buf[cnt] = 0;
 501
 502         ret = strict_strtoul(buf, 10, &val);
 503         if (ret < 0)
 504                 return ret;
 505
 506         ret = tracing_update_buffers();
 507         if (ret < 0)
 508                 return ret;
 509
 510         if (val != 0 && val != 1)
 511                 return -EINVAL;
 512
 513         ret = __ftrace_set_clr_event(NULL, system, NULL, val);
 514         if (ret)
 515                 goto out;
 516
 517         ret = cnt;
 518
 519 out:
 520         *ppos += cnt;
 521
 522         return ret;
 523 }
 524
 525 extern char *__bad_type_size(void);
 526
 527 #undef FIELD
 528 #define FIELD(type, name)                                               \
 529         sizeof(type) != sizeof(field.name) ? __bad_type_size() :        \
 530         #type, "common_" #name, offsetof(typeof(field), name),          \
 531                 sizeof(field.name)
 532
 533 static int trace_write_header(struct trace_seq *s)
 534 {
 535         struct trace_entry field;
 536
 537         /* struct trace_entry */
 538         return trace_seq_printf(s,
 539                                 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
 540                                 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
 541                                 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
 542                                 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
 543                                 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
 544                                 "\n",
 545                                 FIELD(unsigned short, type),
 546                                 FIELD(unsigned char, flags),
 547                                 FIELD(unsigned char, preempt_count),
 548                                 FIELD(int, pid),
 549                                 FIELD(int, tgid));
 550 }
 551
 552 static ssize_t
 553 event_format_read(struct file *filp, char __user *ubuf, size_t cnt,
 554                   loff_t *ppos)
 555 {
 556         struct ftrace_event_call *call = filp->private_data;
 557         struct trace_seq *s;
 558         char *buf;
 559         int r;
 560
 561         if (*ppos)
 562                 return 0;
 563
 564         s = kmalloc(sizeof(*s), GFP_KERNEL);
 565         if (!s)
 566                 return -ENOMEM;
 567
 568         trace_seq_init(s);
 569
 570         /* If any of the first writes fail, so will the show_format. */
 571
 572         trace_seq_printf(s, "name: %s\n", call->name);
 573         trace_seq_printf(s, "ID: %d\n", call->id);
 574         trace_seq_printf(s, "format:\n");
 575         trace_write_header(s);
 576
 577         r = call->show_format(s);
 578         if (!r) {
 579                 /*
 580                  * ug!  The format output is bigger than a PAGE!!
 581                  */
 582                 buf = "FORMAT TOO BIG\n";
 583                 r = simple_read_from_buffer(ubuf, cnt, ppos,
 584                                               buf, strlen(buf));
 585                 goto out;
 586         }
 587
 588         r = simple_read_from_buffer(ubuf, cnt, ppos,
 589                                     s->buffer, s->len);
 590  out:
 591         kfree(s);
 592         return r;
 593 }
 594
 595 static ssize_t
 596 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 597 {
 598         struct ftrace_event_call *call = filp->private_data;
 599         struct trace_seq *s;
 600         int r;
 601
 602         if (*ppos)
 603                 return 0;
 604
 605         s = kmalloc(sizeof(*s), GFP_KERNEL);
 606         if (!s)
 607                 return -ENOMEM;
 608
 609         trace_seq_init(s);
 610         trace_seq_printf(s, "%d\n", call->id);
 611
 612         r = simple_read_from_buffer(ubuf, cnt, ppos,
 613                                     s->buffer, s->len);
 614         kfree(s);
 615         return r;
 616 }
 617
 618 static ssize_t
 619 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 620                   loff_t *ppos)
 621 {
 622         struct ftrace_event_call *call = filp->private_data;
 623         struct trace_seq *s;
 624         int r;
 625
 626         if (*ppos)
 627                 return 0;
 628
 629         s = kmalloc(sizeof(*s), GFP_KERNEL);
 630         if (!s)
 631                 return -ENOMEM;
 632
 633         trace_seq_init(s);
 634
 635         print_event_filter(call, s);
 636         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 637
 638         kfree(s);
 639
 640         return r;
 641 }
 642
 643 static ssize_t
 644 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
 645                    loff_t *ppos)
 646 {
 647         struct ftrace_event_call *call = filp->private_data;
 648         char *buf;
 649         int err;
 650
 651         if (cnt >= PAGE_SIZE)
 652                 return -EINVAL;
 653
 654         buf = (char *)__get_free_page(GFP_TEMPORARY);
 655         if (!buf)
 656                 return -ENOMEM;
 657
 658         if (copy_from_user(buf, ubuf, cnt)) {
 659                 free_page((unsigned long) buf);
 660                 return -EFAULT;
 661         }
 662         buf[cnt] = '\0';
 663
 664         err = apply_event_filter(call, buf);
 665         free_page((unsigned long) buf);
 666         if (err < 0)
 667                 return err;
 668
 669         *ppos += cnt;
 670
 671         return cnt;
 672 }
 673
 674 static ssize_t
 675 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 676                       loff_t *ppos)
 677 {
 678         struct event_subsystem *system = filp->private_data;
 679         struct trace_seq *s;
 680         int r;
 681
 682         if (*ppos)
 683                 return 0;
 684
 685         s = kmalloc(sizeof(*s), GFP_KERNEL);
 686         if (!s)
 687                 return -ENOMEM;
 688
 689         trace_seq_init(s);
 690
 691         print_subsystem_event_filter(system, s);
 692         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 693
 694         kfree(s);
 695
 696         return r;
 697 }
 698
 699 static ssize_t
 700 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
 701                        loff_t *ppos)
 702 {
 703         struct event_subsystem *system = filp->private_data;
 704         char *buf;
 705         int err;
 706
 707         if (cnt >= PAGE_SIZE)
 708                 return -EINVAL;
 709
 710         buf = (char *)__get_free_page(GFP_TEMPORARY);
 711         if (!buf)
 712                 return -ENOMEM;
 713
 714         if (copy_from_user(buf, ubuf, cnt)) {
 715                 free_page((unsigned long) buf);
 716                 return -EFAULT;
 717         }
 718         buf[cnt] = '\0';
 719
 720         err = apply_subsystem_event_filter(system, buf);
 721         free_page((unsigned long) buf);
 722         if (err < 0)
 723                 return err;
 724
 725         *ppos += cnt;
 726
 727         return cnt;
 728 }
 729
 730 static ssize_t
 731 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 732 {
 733         int (*func)(struct trace_seq *s) = filp->private_data;
 734         struct trace_seq *s;
 735         int r;
 736
 737         if (*ppos)
 738                 return 0;
 739
 740         s = kmalloc(sizeof(*s), GFP_KERNEL);
 741         if (!s)
 742                 return -ENOMEM;
 743
 744         trace_seq_init(s);
 745
 746         func(s);
 747         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 748
 749         kfree(s);
 750
 751         return r;
 752 }
 753
 754 static const struct seq_operations show_event_seq_ops = {
 755         .start = t_start,
 756         .next = t_next,
 757         .show = t_show,
 758         .stop = t_stop,
 759 };
 760
 761 static const struct seq_operations show_set_event_seq_ops = {
 762         .start = s_start,
 763         .next = s_next,
 764         .show = t_show,
 765         .stop = t_stop,
 766 };
 767
 768 static const struct file_operations ftrace_avail_fops = {
 769         .open = ftrace_event_seq_open,
 770         .read = seq_read,
 771         .llseek = seq_lseek,
 772         .release = seq_release,
 773 };
 774
 775 static const struct file_operations ftrace_set_event_fops = {
 776         .open = ftrace_event_seq_open,
 777         .read = seq_read,
 778         .write = ftrace_event_write,
 779         .llseek = seq_lseek,
 780         .release = seq_release,
 781 };
 782
 783 static const struct file_operations ftrace_enable_fops = {
 784         .open = tracing_open_generic,
 785         .read = event_enable_read,
 786         .write = event_enable_write,
 787 };
 788
 789 static const struct file_operations ftrace_event_format_fops = {
 790         .open = tracing_open_generic,
 791         .read = event_format_read,
 792 };
 793
 794 static const struct file_operations ftrace_event_id_fops = {
 795         .open = tracing_open_generic,
 796         .read = event_id_read,
 797 };
 798
 799 static const struct file_operations ftrace_event_filter_fops = {
 800         .open = tracing_open_generic,
 801         .read = event_filter_read,
 802         .write = event_filter_write,
 803 };
 804
 805 static const struct file_operations ftrace_subsystem_filter_fops = {
 806         .open = tracing_open_generic,
 807         .read = subsystem_filter_read,
 808         .write = subsystem_filter_write,
 809 };
 810
 811 static const struct file_operations ftrace_system_enable_fops = {
 812         .open = tracing_open_generic,
 813         .read = system_enable_read,
 814         .write = system_enable_write,
 815 };
 816
 817 static const struct file_operations ftrace_show_header_fops = {
 818         .open = tracing_open_generic,
 819         .read = show_header,
 820 };
 821
 822 static struct dentry *event_trace_events_dir(void)
 823 {
 824         static struct dentry *d_tracer;
 825         static struct dentry *d_events;
 826
 827         if (d_events)
 828                 return d_events;
 829
 830         d_tracer = tracing_init_dentry();
 831         if (!d_tracer)
 832                 return NULL;
 833
 834         d_events = debugfs_create_dir("events", d_tracer);
 835         if (!d_events)
 836                 pr_warning("Could not create debugfs "
 837                            "'events' directory\n");
 838
 839         return d_events;
 840 }
 841
 842 static LIST_HEAD(event_subsystems);
 843
 844 static struct dentry *
 845 event_subsystem_dir(const char *name, struct dentry *d_events)
 846 {
 847         struct event_subsystem *system;
 848         struct dentry *entry;
 849
 850         /* First see if we did not already create this dir */
 851         list_for_each_entry(system, &event_subsystems, list) {
 852                 if (strcmp(system->name, name) == 0)
 853                         return system->entry;
 854         }
 855
 856         /* need to create new entry */
 857         system = kmalloc(sizeof(*system), GFP_KERNEL);
 858         if (!system) {
 859                 pr_warning("No memory to create event subsystem %s\n",
 860                            name);
 861                 return d_events;
 862         }
 863
 864         system->entry = debugfs_create_dir(name, d_events);
 865         if (!system->entry) {
 866                 pr_warning("Could not create event subsystem %s\n",
 867                            name);
 868                 kfree(system);
 869                 return d_events;
 870         }
 871
 872         system->name = kstrdup(name, GFP_KERNEL);
 873         if (!system->name) {
 874                 debugfs_remove(system->entry);
 875                 kfree(system);
 876                 return d_events;
 877         }
 878
 879         list_add(&system->list, &event_subsystems);
 880
 881         system->filter = NULL;
 882
 883         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
 884         if (!system->filter) {
 885                 pr_warning("Could not allocate filter for subsystem "
 886                            "'%s'\n", name);
 887                 return system->entry;
 888         }
 889
 890         entry = debugfs_create_file("filter", 0644, system->entry, system,
 891                                     &ftrace_subsystem_filter_fops);
 892         if (!entry) {
 893                 kfree(system->filter);
 894                 system->filter = NULL;
 895                 pr_warning("Could not create debugfs "
 896                            "'%s/filter' entry\n", name);
 897         }
 898
 899         entry = trace_create_file("enable", 0644, system->entry,
 900                                   (void *)system->name,
 901                                   &ftrace_system_enable_fops);
 902
 903         return system->entry;
 904 }
 905
 906 static int
 907 event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
 908                  const struct file_operations *id,
 909                  const struct file_operations *enable,
 910                  const struct file_operations *filter,
 911                  const struct file_operations *format)
 912 {
 913         struct dentry *entry;
 914         int ret;
 915
 916         /*
 917          * If the trace point header did not define TRACE_SYSTEM
 918          * then the system would be called "TRACE_SYSTEM".
 919          */
 920         if (strcmp(call->system, TRACE_SYSTEM) != 0)
 921                 d_events = event_subsystem_dir(call->system, d_events);
 922
 923         if (call->raw_init) {
 924                 ret = call->raw_init();
 925                 if (ret < 0) {
 926                         pr_warning("Could not initialize trace point"
 927                                    " events/%s\n", call->name);
 928                         return ret;
 929                 }
 930         }
 931
 932         call->dir = debugfs_create_dir(call->name, d_events);
 933         if (!call->dir) {
 934                 pr_warning("Could not create debugfs "
 935                            "'%s' directory\n", call->name);
 936                 return -1;
 937         }
 938
 939         if (call->regfunc)
 940                 entry = trace_create_file("enable", 0644, call->dir, call,
 941                                           enable);
 942
 943         if (call->id)
 944                 entry = trace_create_file("id", 0444, call->dir, call,
 945                                           id);
 946
 947         if (call->define_fields) {
 948                 ret = call->define_fields();
 949                 if (ret < 0) {
 950                         pr_warning("Could not initialize trace point"
 951                                    " events/%s\n", call->name);
 952                         return ret;
 953                 }
 954                 entry = trace_create_file("filter", 0644, call->dir, call,
 955                                           filter);
 956         }
 957
 958         /* A trace may not want to export its format */
 959         if (!call->show_format)
 960                 return 0;
 961
 962         entry = trace_create_file("format", 0444, call->dir, call,
 963                                   format);
 964
 965         return 0;
 966 }
 967
 968 #define for_each_event(event, start, end)                       \
 969         for (event = start;                                     \
 970              (unsigned long)event < (unsigned long)end;         \
 971              event++)
 972
 973 #ifdef CONFIG_MODULES
 974
 975 static LIST_HEAD(ftrace_module_file_list);
 976
 977 /*
 978  * Modules must own their file_operations to keep up with
 979  * reference counting.
 980  */
 981 struct ftrace_module_file_ops {
 982         struct list_head                list;
 983         struct module                   *mod;
 984         struct file_operations          id;
 985         struct file_operations          enable;
 986         struct file_operations          format;
 987         struct file_operations          filter;
 988 };
 989
 990 static struct ftrace_module_file_ops *
 991 trace_create_file_ops(struct module *mod)
 992 {
 993         struct ftrace_module_file_ops *file_ops;
 994
 995         /*
 996          * This is a bit of a PITA. To allow for correct reference
 997          * counting, modules must "own" their file_operations.
 998          * To do this, we allocate the file operations that will be
 999          * used in the event directory.
1000          */
1001
1002         file_ops = kmalloc(sizeof(*file_ops), GFP_KERNEL);
1003         if (!file_ops)
1004                 return NULL;
1005
1006         file_ops->mod = mod;
1007
1008         file_ops->id = ftrace_event_id_fops;
1009         file_ops->id.owner = mod;
1010
1011         file_ops->enable = ftrace_enable_fops;
1012         file_ops->enable.owner = mod;
1013
1014         file_ops->filter = ftrace_event_filter_fops;
1015         file_ops->filter.owner = mod;
1016
1017         file_ops->format = ftrace_event_format_fops;
1018         file_ops->format.owner = mod;
1019
1020         list_add(&file_ops->list, &ftrace_module_file_list);
1021
1022         return file_ops;
1023 }
1024
1025 static void trace_module_add_events(struct module *mod)
1026 {
1027         struct ftrace_module_file_ops *file_ops = NULL;
1028         struct ftrace_event_call *call, *start, *end;
1029         struct dentry *d_events;
1030
1031         start = mod->trace_events;
1032         end = mod->trace_events + mod->num_trace_events;
1033
1034         if (start == end)
1035                 return;
1036
1037         d_events = event_trace_events_dir();
1038         if (!d_events)
1039                 return;
1040
1041         for_each_event(call, start, end) {
1042                 /* The linker may leave blanks */
1043                 if (!call->name)
1044                         continue;
1045
1046                 /*
1047                  * This module has events, create file ops for this module
1048                  * if not already done.
1049                  */
1050                 if (!file_ops) {
1051                         file_ops = trace_create_file_ops(mod);
1052                         if (!file_ops)
1053                                 return;
1054                 }
1055                 call->mod = mod;
1056                 list_add(&call->list, &ftrace_events);
1057                 event_create_dir(call, d_events,
1058                                  &file_ops->id, &file_ops->enable,
1059                                  &file_ops->filter, &file_ops->format);
1060         }
1061 }
1062
1063 static void trace_module_remove_events(struct module *mod)
1064 {
1065         struct ftrace_module_file_ops *file_ops;
1066         struct ftrace_event_call *call, *p;
1067         bool found = false;
1068
1069         down_write(&trace_event_mutex);
1070         list_for_each_entry_safe(call, p, &ftrace_events, list) {
1071                 if (call->mod == mod) {
1072                         found = true;
1073                         ftrace_event_enable_disable(call, 0);
1074                         if (call->event)
1075                                 __unregister_ftrace_event(call->event);
1076                         debugfs_remove_recursive(call->dir);
1077                         list_del(&call->list);
1078                         trace_destroy_fields(call);
1079                         destroy_preds(call);
1080                 }
1081         }
1082
1083         /* Now free the file_operations */
1084         list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1085                 if (file_ops->mod == mod)
1086                         break;
1087         }
1088         if (&file_ops->list != &ftrace_module_file_list) {
1089                 list_del(&file_ops->list);
1090                 kfree(file_ops);
1091         }
1092
1093         /*
1094          * It is safest to reset the ring buffer if the module being unloaded
1095          * registered any events.
1096          */
1097         if (found)
1098                 tracing_reset_current_online_cpus();
1099         up_write(&trace_event_mutex);
1100 }
1101
1102 static int trace_module_notify(struct notifier_block *self,
1103                                unsigned long val, void *data)
1104 {
1105         struct module *mod = data;
1106
1107         mutex_lock(&event_mutex);
1108         switch (val) {
1109         case MODULE_STATE_COMING:
1110                 trace_module_add_events(mod);
1111                 break;
1112         case MODULE_STATE_GOING:
1113                 trace_module_remove_events(mod);
1114                 break;
1115         }
1116         mutex_unlock(&event_mutex);
1117
1118         return 0;
1119 }
1120 #else
1121 static int trace_module_notify(struct notifier_block *self,
1122                                unsigned long val, void *data)
1123 {
1124         return 0;
1125 }
1126 #endif /* CONFIG_MODULES */
1127
1128 struct notifier_block trace_module_nb = {
1129         .notifier_call = trace_module_notify,
1130         .priority = 0,
1131 };
1132
1133 extern struct ftrace_event_call __start_ftrace_events[];
1134 extern struct ftrace_event_call __stop_ftrace_events[];
1135
1136 static __init int event_trace_init(void)
1137 {
1138         struct ftrace_event_call *call;
1139         struct dentry *d_tracer;
1140         struct dentry *entry;
1141         struct dentry *d_events;
1142         int ret;
1143
1144         d_tracer = tracing_init_dentry();
1145         if (!d_tracer)
1146                 return 0;
1147
1148         entry = debugfs_create_file("available_events", 0444, d_tracer,
1149                                     (void *)&show_event_seq_ops,
1150                                     &ftrace_avail_fops);
1151         if (!entry)
1152                 pr_warning("Could not create debugfs "
1153                            "'available_events' entry\n");
1154
1155         entry = debugfs_create_file("set_event", 0644, d_tracer,
1156                                     (void *)&show_set_event_seq_ops,
1157                                     &ftrace_set_event_fops);
1158         if (!entry)
1159                 pr_warning("Could not create debugfs "
1160                            "'set_event' entry\n");
1161
1162         d_events = event_trace_events_dir();
1163         if (!d_events)
1164                 return 0;
1165
1166         /* ring buffer internal formats */
1167         trace_create_file("header_page", 0444, d_events,
1168                           ring_buffer_print_page_header,
1169                           &ftrace_show_header_fops);
1170
1171         trace_create_file("header_event", 0444, d_events,
1172                           ring_buffer_print_entry_header,
1173                           &ftrace_show_header_fops);
1174
1175         trace_create_file("enable", 0644, d_events,
1176                           NULL, &ftrace_system_enable_fops);
1177
1178         for_each_event(call, __start_ftrace_events, __stop_ftrace_events) {
1179                 /* The linker may leave blanks */
1180                 if (!call->name)
1181                         continue;
1182                 list_add(&call->list, &ftrace_events);
1183                 event_create_dir(call, d_events, &ftrace_event_id_fops,
1184                                  &ftrace_enable_fops, &ftrace_event_filter_fops,
1185                                  &ftrace_event_format_fops);
1186         }
1187
1188         ret = register_module_notifier(&trace_module_nb);
1189         if (ret)
1190                 pr_warning("Failed to register trace events module notifier\n");
1191
1192         return 0;
1193 }
1194 fs_initcall(event_trace_init);
1195
1196 #ifdef CONFIG_FTRACE_STARTUP_TEST
1197
1198 static DEFINE_SPINLOCK(test_spinlock);
1199 static DEFINE_SPINLOCK(test_spinlock_irq);
1200 static DEFINE_MUTEX(test_mutex);
1201
1202 static __init void test_work(struct work_struct *dummy)
1203 {
1204         spin_lock(&test_spinlock);
1205         spin_lock_irq(&test_spinlock_irq);
1206         udelay(1);
1207         spin_unlock_irq(&test_spinlock_irq);
1208         spin_unlock(&test_spinlock);
1209
1210         mutex_lock(&test_mutex);
1211         msleep(1);
1212         mutex_unlock(&test_mutex);
1213 }
1214
1215 static __init int event_test_thread(void *unused)
1216 {
1217         void *test_malloc;
1218
1219         test_malloc = kmalloc(1234, GFP_KERNEL);
1220         if (!test_malloc)
1221                 pr_info("failed to kmalloc\n");
1222
1223         schedule_on_each_cpu(test_work);
1224
1225         kfree(test_malloc);
1226
1227         set_current_state(TASK_INTERRUPTIBLE);
1228         while (!kthread_should_stop())
1229                 schedule();
1230
1231         return 0;
1232 }
1233
1234 /*
1235  * Do various things that may trigger events.
1236  */
1237 static __init void event_test_stuff(void)
1238 {
1239         struct task_struct *test_thread;
1240
1241         test_thread = kthread_run(event_test_thread, NULL, "test-events");
1242         msleep(1);
1243         kthread_stop(test_thread);
1244 }
1245
1246 /*
1247  * For every trace event defined, we will test each trace point separately,
1248  * and then by groups, and finally all trace points.
1249  */
1250 static __init void event_trace_self_tests(void)
1251 {
1252         struct ftrace_event_call *call;
1253         struct event_subsystem *system;
1254         int ret;
1255
1256         pr_info("Running tests on trace events:\n");
1257
1258         list_for_each_entry(call, &ftrace_events, list) {
1259
1260                 /* Only test those that have a regfunc */
1261                 if (!call->regfunc)
1262                         continue;
1263
1264                 pr_info("Testing event %s: ", call->name);
1265
1266                 /*
1267                  * If an event is already enabled, someone is using
1268                  * it and the self test should not be on.
1269                  */
1270                 if (call->enabled) {
1271                         pr_warning("Enabled event during self test!\n");
1272                         WARN_ON_ONCE(1);
1273                         continue;
1274                 }
1275
1276                 ftrace_event_enable_disable(call, 1);
1277                 event_test_stuff();
1278                 ftrace_event_enable_disable(call, 0);
1279
1280                 pr_cont("OK\n");
1281         }
1282
1283         /* Now test at the sub system level */
1284
1285         pr_info("Running tests on trace event systems:\n");
1286
1287         list_for_each_entry(system, &event_subsystems, list) {
1288
1289                 /* the ftrace system is special, skip it */
1290                 if (strcmp(system->name, "ftrace") == 0)
1291                         continue;
1292
1293                 pr_info("Testing event system %s: ", system->name);
1294
1295                 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 1);
1296                 if (WARN_ON_ONCE(ret)) {
1297                         pr_warning("error enabling system %s\n",
1298                                    system->name);
1299                         continue;
1300                 }
1301
1302                 event_test_stuff();
1303
1304                 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 0);
1305                 if (WARN_ON_ONCE(ret))
1306                         pr_warning("error disabling system %s\n",
1307                                    system->name);
1308
1309                 pr_cont("OK\n");
1310         }
1311
1312         /* Test with all events enabled */
1313
1314         pr_info("Running tests on all trace events:\n");
1315         pr_info("Testing all events: ");
1316
1317         ret = __ftrace_set_clr_event(NULL, NULL, NULL, 1);
1318         if (WARN_ON_ONCE(ret)) {
1319                 pr_warning("error enabling all events\n");
1320                 return;
1321         }
1322
1323         event_test_stuff();
1324
1325         /* reset sysname */
1326         ret = __ftrace_set_clr_event(NULL, NULL, NULL, 0);
1327         if (WARN_ON_ONCE(ret)) {
1328                 pr_warning("error disabling all events\n");
1329                 return;
1330         }
1331
1332         pr_cont("OK\n");
1333 }
1334
1335 #ifdef CONFIG_FUNCTION_TRACER
1336
1337 static DEFINE_PER_CPU(atomic_t, test_event_disable);
1338
1339 static void
1340 function_test_events_call(unsigned long ip, unsigned long parent_ip)
1341 {
1342         struct ring_buffer_event *event;
1343         struct ftrace_entry *entry;
1344         unsigned long flags;
1345         long disabled;
1346         int resched;
1347         int cpu;
1348         int pc;
1349
1350         pc = preempt_count();
1351         resched = ftrace_preempt_disable();
1352         cpu = raw_smp_processor_id();
1353         disabled = atomic_inc_return(&per_cpu(test_event_disable, cpu));
1354
1355         if (disabled != 1)
1356                 goto out;
1357
1358         local_save_flags(flags);
1359
1360         event = trace_current_buffer_lock_reserve(TRACE_FN, sizeof(*entry),
1361                                                   flags, pc);
1362         if (!event)
1363                 goto out;
1364         entry   = ring_buffer_event_data(event);
1365         entry->ip                       = ip;
1366         entry->parent_ip                = parent_ip;
1367
1368         trace_nowake_buffer_unlock_commit(event, flags, pc);
1369
1370  out:
1371         atomic_dec(&per_cpu(test_event_disable, cpu));
1372         ftrace_preempt_enable(resched);
1373 }
1374
1375 static struct ftrace_ops trace_ops __initdata  =
1376 {
1377         .func = function_test_events_call,
1378 };
1379
1380 static __init void event_trace_self_test_with_function(void)
1381 {
1382         register_ftrace_function(&trace_ops);
1383         pr_info("Running tests again, along with the function tracer\n");
1384         event_trace_self_tests();
1385         unregister_ftrace_function(&trace_ops);
1386 }
1387 #else
1388 static __init void event_trace_self_test_with_function(void)
1389 {
1390 }
1391 #endif
1392
1393 static __init int event_trace_self_tests_init(void)
1394 {
1395
1396         event_trace_self_tests();
1397
1398         event_trace_self_test_with_function();
1399
1400         return 0;
1401 }
1402
1403 late_initcall(event_trace_self_tests_init);
1404
1405 #endif