SAFE public projects git trees. - safe/jmp/linux-2.6/blob - kernel/trace/trace_events.c

   1 /*
   2  * event tracer
   3  *
   4  * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
   5  *
   6  *  - Added format output of fields of the trace point.
   7  *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
   8  *
   9  */
  10
  11 #include <linux/workqueue.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/kthread.h>
  14 #include <linux/debugfs.h>
  15 #include <linux/uaccess.h>
  16 #include <linux/module.h>
  17 #include <linux/ctype.h>
  18 #include <linux/delay.h>
  19
  20 #include <asm/setup.h>
  21
  22 #include "trace_output.h"
  23
  24 #define TRACE_SYSTEM "TRACE_SYSTEM"
  25
  26 DEFINE_MUTEX(event_mutex);
  27
  28 LIST_HEAD(ftrace_events);
  29
  30 int trace_define_field(struct ftrace_event_call *call, const char *type,
  31                        const char *name, int offset, int size, int is_signed)
  32 {
  33         struct ftrace_event_field *field;
  34
  35         field = kzalloc(sizeof(*field), GFP_KERNEL);
  36         if (!field)
  37                 goto err;
  38
  39         field->name = kstrdup(name, GFP_KERNEL);
  40         if (!field->name)
  41                 goto err;
  42
  43         field->type = kstrdup(type, GFP_KERNEL);
  44         if (!field->type)
  45                 goto err;
  46
  47         field->filter_type = filter_assign_type(type);
  48         field->offset = offset;
  49         field->size = size;
  50         field->is_signed = is_signed;
  51
  52         list_add(&field->link, &call->fields);
  53
  54         return 0;
  55
  56 err:
  57         if (field) {
  58                 kfree(field->name);
  59                 kfree(field->type);
  60         }
  61         kfree(field);
  62
  63         return -ENOMEM;
  64 }
  65 EXPORT_SYMBOL_GPL(trace_define_field);
  66
  67 #define __common_field(type, item)                                      \
  68         ret = trace_define_field(call, #type, "common_" #item,          \
  69                                  offsetof(typeof(ent), item),           \
  70                                  sizeof(ent.item),                      \
  71                                  is_signed_type(type));                 \
  72         if (ret)                                                        \
  73                 return ret;
  74
  75 int trace_define_common_fields(struct ftrace_event_call *call)
  76 {
  77         int ret;
  78         struct trace_entry ent;
  79
  80         __common_field(unsigned short, type);
  81         __common_field(unsigned char, flags);
  82         __common_field(unsigned char, preempt_count);
  83         __common_field(int, pid);
  84         __common_field(int, tgid);
  85
  86         return ret;
  87 }
  88 EXPORT_SYMBOL_GPL(trace_define_common_fields);
  89
  90 #ifdef CONFIG_MODULES
  91
  92 static void trace_destroy_fields(struct ftrace_event_call *call)
  93 {
  94         struct ftrace_event_field *field, *next;
  95
  96         list_for_each_entry_safe(field, next, &call->fields, link) {
  97                 list_del(&field->link);
  98                 kfree(field->type);
  99                 kfree(field->name);
 100                 kfree(field);
 101         }
 102 }
 103
 104 #endif /* CONFIG_MODULES */
 105
 106 static void ftrace_event_enable_disable(struct ftrace_event_call *call,
 107                                         int enable)
 108 {
 109         switch (enable) {
 110         case 0:
 111                 if (call->enabled) {
 112                         call->enabled = 0;
 113                         tracing_stop_cmdline_record();
 114                         call->unregfunc(call->data);
 115                 }
 116                 break;
 117         case 1:
 118                 if (!call->enabled) {
 119                         call->enabled = 1;
 120                         tracing_start_cmdline_record();
 121                         call->regfunc(call->data);
 122                 }
 123                 break;
 124         }
 125 }
 126
 127 static void ftrace_clear_events(void)
 128 {
 129         struct ftrace_event_call *call;
 130
 131         mutex_lock(&event_mutex);
 132         list_for_each_entry(call, &ftrace_events, list) {
 133                 ftrace_event_enable_disable(call, 0);
 134         }
 135         mutex_unlock(&event_mutex);
 136 }
 137
 138 /*
 139  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 140  */
 141 static int __ftrace_set_clr_event(const char *match, const char *sub,
 142                                   const char *event, int set)
 143 {
 144         struct ftrace_event_call *call;
 145         int ret = -EINVAL;
 146
 147         mutex_lock(&event_mutex);
 148         list_for_each_entry(call, &ftrace_events, list) {
 149
 150                 if (!call->name || !call->regfunc)
 151                         continue;
 152
 153                 if (match &&
 154                     strcmp(match, call->name) != 0 &&
 155                     strcmp(match, call->system) != 0)
 156                         continue;
 157
 158                 if (sub && strcmp(sub, call->system) != 0)
 159                         continue;
 160
 161                 if (event && strcmp(event, call->name) != 0)
 162                         continue;
 163
 164                 ftrace_event_enable_disable(call, set);
 165
 166                 ret = 0;
 167         }
 168         mutex_unlock(&event_mutex);
 169
 170         return ret;
 171 }
 172
 173 static int ftrace_set_clr_event(char *buf, int set)
 174 {
 175         char *event = NULL, *sub = NULL, *match;
 176
 177         /*
 178          * The buf format can be <subsystem>:<event-name>
 179          *  *:<event-name> means any event by that name.
 180          *  :<event-name> is the same.
 181          *
 182          *  <subsystem>:* means all events in that subsystem
 183          *  <subsystem>: means the same.
 184          *
 185          *  <name> (no ':') means all events in a subsystem with
 186          *  the name <name> or any event that matches <name>
 187          */
 188
 189         match = strsep(&buf, ":");
 190         if (buf) {
 191                 sub = match;
 192                 event = buf;
 193                 match = NULL;
 194
 195                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 196                         sub = NULL;
 197                 if (!strlen(event) || strcmp(event, "*") == 0)
 198                         event = NULL;
 199         }
 200
 201         return __ftrace_set_clr_event(match, sub, event, set);
 202 }
 203
 204 /**
 205  * trace_set_clr_event - enable or disable an event
 206  * @system: system name to match (NULL for any system)
 207  * @event: event name to match (NULL for all events, within system)
 208  * @set: 1 to enable, 0 to disable
 209  *
 210  * This is a way for other parts of the kernel to enable or disable
 211  * event recording.
 212  *
 213  * Returns 0 on success, -EINVAL if the parameters do not match any
 214  * registered events.
 215  */
 216 int trace_set_clr_event(const char *system, const char *event, int set)
 217 {
 218         return __ftrace_set_clr_event(NULL, system, event, set);
 219 }
 220
 221 /* 128 should be much more than enough */
 222 #define EVENT_BUF_SIZE          127
 223
 224 static ssize_t
 225 ftrace_event_write(struct file *file, const char __user *ubuf,
 226                    size_t cnt, loff_t *ppos)
 227 {
 228         size_t read = 0;
 229         int i, set = 1;
 230         ssize_t ret;
 231         char *buf;
 232         char ch;
 233
 234         if (!cnt || cnt < 0)
 235                 return 0;
 236
 237         ret = tracing_update_buffers();
 238         if (ret < 0)
 239                 return ret;
 240
 241         ret = get_user(ch, ubuf++);
 242         if (ret)
 243                 return ret;
 244         read++;
 245         cnt--;
 246
 247         /* skip white space */
 248         while (cnt && isspace(ch)) {
 249                 ret = get_user(ch, ubuf++);
 250                 if (ret)
 251                         return ret;
 252                 read++;
 253                 cnt--;
 254         }
 255
 256         /* Only white space found? */
 257         if (isspace(ch)) {
 258                 file->f_pos += read;
 259                 ret = read;
 260                 return ret;
 261         }
 262
 263         buf = kmalloc(EVENT_BUF_SIZE+1, GFP_KERNEL);
 264         if (!buf)
 265                 return -ENOMEM;
 266
 267         if (cnt > EVENT_BUF_SIZE)
 268                 cnt = EVENT_BUF_SIZE;
 269
 270         i = 0;
 271         while (cnt && !isspace(ch)) {
 272                 if (!i && ch == '!')
 273                         set = 0;
 274                 else
 275                         buf[i++] = ch;
 276
 277                 ret = get_user(ch, ubuf++);
 278                 if (ret)
 279                         goto out_free;
 280                 read++;
 281                 cnt--;
 282         }
 283         buf[i] = 0;
 284
 285         file->f_pos += read;
 286
 287         ret = ftrace_set_clr_event(buf, set);
 288         if (ret)
 289                 goto out_free;
 290
 291         ret = read;
 292
 293  out_free:
 294         kfree(buf);
 295
 296         return ret;
 297 }
 298
 299 static void *
 300 t_next(struct seq_file *m, void *v, loff_t *pos)
 301 {
 302         struct list_head *list = m->private;
 303         struct ftrace_event_call *call;
 304
 305         (*pos)++;
 306
 307         for (;;) {
 308                 if (list == &ftrace_events)
 309                         return NULL;
 310
 311                 call = list_entry(list, struct ftrace_event_call, list);
 312
 313                 /*
 314                  * The ftrace subsystem is for showing formats only.
 315                  * They can not be enabled or disabled via the event files.
 316                  */
 317                 if (call->regfunc)
 318                         break;
 319
 320                 list = list->next;
 321         }
 322
 323         m->private = list->next;
 324
 325         return call;
 326 }
 327
 328 static void *t_start(struct seq_file *m, loff_t *pos)
 329 {
 330         struct ftrace_event_call *call = NULL;
 331         loff_t l;
 332
 333         mutex_lock(&event_mutex);
 334
 335         m->private = ftrace_events.next;
 336         for (l = 0; l <= *pos; ) {
 337                 call = t_next(m, NULL, &l);
 338                 if (!call)
 339                         break;
 340         }
 341         return call;
 342 }
 343
 344 static void *
 345 s_next(struct seq_file *m, void *v, loff_t *pos)
 346 {
 347         struct list_head *list = m->private;
 348         struct ftrace_event_call *call;
 349
 350         (*pos)++;
 351
 352  retry:
 353         if (list == &ftrace_events)
 354                 return NULL;
 355
 356         call = list_entry(list, struct ftrace_event_call, list);
 357
 358         if (!call->enabled) {
 359                 list = list->next;
 360                 goto retry;
 361         }
 362
 363         m->private = list->next;
 364
 365         return call;
 366 }
 367
 368 static void *s_start(struct seq_file *m, loff_t *pos)
 369 {
 370         struct ftrace_event_call *call = NULL;
 371         loff_t l;
 372
 373         mutex_lock(&event_mutex);
 374
 375         m->private = ftrace_events.next;
 376         for (l = 0; l <= *pos; ) {
 377                 call = s_next(m, NULL, &l);
 378                 if (!call)
 379                         break;
 380         }
 381         return call;
 382 }
 383
 384 static int t_show(struct seq_file *m, void *v)
 385 {
 386         struct ftrace_event_call *call = v;
 387
 388         if (strcmp(call->system, TRACE_SYSTEM) != 0)
 389                 seq_printf(m, "%s:", call->system);
 390         seq_printf(m, "%s\n", call->name);
 391
 392         return 0;
 393 }
 394
 395 static void t_stop(struct seq_file *m, void *p)
 396 {
 397         mutex_unlock(&event_mutex);
 398 }
 399
 400 static int
 401 ftrace_event_seq_open(struct inode *inode, struct file *file)
 402 {
 403         const struct seq_operations *seq_ops;
 404
 405         if ((file->f_mode & FMODE_WRITE) &&
 406             (file->f_flags & O_TRUNC))
 407                 ftrace_clear_events();
 408
 409         seq_ops = inode->i_private;
 410         return seq_open(file, seq_ops);
 411 }
 412
 413 static ssize_t
 414 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 415                   loff_t *ppos)
 416 {
 417         struct ftrace_event_call *call = filp->private_data;
 418         char *buf;
 419
 420         if (call->enabled)
 421                 buf = "1\n";
 422         else
 423                 buf = "0\n";
 424
 425         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 426 }
 427
 428 static ssize_t
 429 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 430                    loff_t *ppos)
 431 {
 432         struct ftrace_event_call *call = filp->private_data;
 433         char buf[64];
 434         unsigned long val;
 435         int ret;
 436
 437         if (cnt >= sizeof(buf))
 438                 return -EINVAL;
 439
 440         if (copy_from_user(&buf, ubuf, cnt))
 441                 return -EFAULT;
 442
 443         buf[cnt] = 0;
 444
 445         ret = strict_strtoul(buf, 10, &val);
 446         if (ret < 0)
 447                 return ret;
 448
 449         ret = tracing_update_buffers();
 450         if (ret < 0)
 451                 return ret;
 452
 453         switch (val) {
 454         case 0:
 455         case 1:
 456                 mutex_lock(&event_mutex);
 457                 ftrace_event_enable_disable(call, val);
 458                 mutex_unlock(&event_mutex);
 459                 break;
 460
 461         default:
 462                 return -EINVAL;
 463         }
 464
 465         *ppos += cnt;
 466
 467         return cnt;
 468 }
 469
 470 static ssize_t
 471 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 472                    loff_t *ppos)
 473 {
 474         const char set_to_char[4] = { '?', '0', '1', 'X' };
 475         const char *system = filp->private_data;
 476         struct ftrace_event_call *call;
 477         char buf[2];
 478         int set = 0;
 479         int ret;
 480
 481         mutex_lock(&event_mutex);
 482         list_for_each_entry(call, &ftrace_events, list) {
 483                 if (!call->name || !call->regfunc)
 484                         continue;
 485
 486                 if (system && strcmp(call->system, system) != 0)
 487                         continue;
 488
 489                 /*
 490                  * We need to find out if all the events are set
 491                  * or if all events or cleared, or if we have
 492                  * a mixture.
 493                  */
 494                 set |= (1 << !!call->enabled);
 495
 496                 /*
 497                  * If we have a mixture, no need to look further.
 498                  */
 499                 if (set == 3)
 500                         break;
 501         }
 502         mutex_unlock(&event_mutex);
 503
 504         buf[0] = set_to_char[set];
 505         buf[1] = '\n';
 506
 507         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 508
 509         return ret;
 510 }
 511
 512 static ssize_t
 513 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 514                     loff_t *ppos)
 515 {
 516         const char *system = filp->private_data;
 517         unsigned long val;
 518         char buf[64];
 519         ssize_t ret;
 520
 521         if (cnt >= sizeof(buf))
 522                 return -EINVAL;
 523
 524         if (copy_from_user(&buf, ubuf, cnt))
 525                 return -EFAULT;
 526
 527         buf[cnt] = 0;
 528
 529         ret = strict_strtoul(buf, 10, &val);
 530         if (ret < 0)
 531                 return ret;
 532
 533         ret = tracing_update_buffers();
 534         if (ret < 0)
 535                 return ret;
 536
 537         if (val != 0 && val != 1)
 538                 return -EINVAL;
 539
 540         ret = __ftrace_set_clr_event(NULL, system, NULL, val);
 541         if (ret)
 542                 goto out;
 543
 544         ret = cnt;
 545
 546 out:
 547         *ppos += cnt;
 548
 549         return ret;
 550 }
 551
 552 extern char *__bad_type_size(void);
 553
 554 #undef FIELD
 555 #define FIELD(type, name)                                               \
 556         sizeof(type) != sizeof(field.name) ? __bad_type_size() :        \
 557         #type, "common_" #name, offsetof(typeof(field), name),          \
 558                 sizeof(field.name)
 559
 560 static int trace_write_header(struct trace_seq *s)
 561 {
 562         struct trace_entry field;
 563
 564         /* struct trace_entry */
 565         return trace_seq_printf(s,
 566                                 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
 567                                 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
 568                                 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
 569                                 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
 570                                 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
 571                                 "\n",
 572                                 FIELD(unsigned short, type),
 573                                 FIELD(unsigned char, flags),
 574                                 FIELD(unsigned char, preempt_count),
 575                                 FIELD(int, pid),
 576                                 FIELD(int, tgid));
 577 }
 578
 579 static ssize_t
 580 event_format_read(struct file *filp, char __user *ubuf, size_t cnt,
 581                   loff_t *ppos)
 582 {
 583         struct ftrace_event_call *call = filp->private_data;
 584         struct trace_seq *s;
 585         char *buf;
 586         int r;
 587
 588         if (*ppos)
 589                 return 0;
 590
 591         s = kmalloc(sizeof(*s), GFP_KERNEL);
 592         if (!s)
 593                 return -ENOMEM;
 594
 595         trace_seq_init(s);
 596
 597         /* If any of the first writes fail, so will the show_format. */
 598
 599         trace_seq_printf(s, "name: %s\n", call->name);
 600         trace_seq_printf(s, "ID: %d\n", call->id);
 601         trace_seq_printf(s, "format:\n");
 602         trace_write_header(s);
 603
 604         r = call->show_format(call, s);
 605         if (!r) {
 606                 /*
 607                  * ug!  The format output is bigger than a PAGE!!
 608                  */
 609                 buf = "FORMAT TOO BIG\n";
 610                 r = simple_read_from_buffer(ubuf, cnt, ppos,
 611                                               buf, strlen(buf));
 612                 goto out;
 613         }
 614
 615         r = simple_read_from_buffer(ubuf, cnt, ppos,
 616                                     s->buffer, s->len);
 617  out:
 618         kfree(s);
 619         return r;
 620 }
 621
 622 static ssize_t
 623 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 624 {
 625         struct ftrace_event_call *call = filp->private_data;
 626         struct trace_seq *s;
 627         int r;
 628
 629         if (*ppos)
 630                 return 0;
 631
 632         s = kmalloc(sizeof(*s), GFP_KERNEL);
 633         if (!s)
 634                 return -ENOMEM;
 635
 636         trace_seq_init(s);
 637         trace_seq_printf(s, "%d\n", call->id);
 638
 639         r = simple_read_from_buffer(ubuf, cnt, ppos,
 640                                     s->buffer, s->len);
 641         kfree(s);
 642         return r;
 643 }
 644
 645 static ssize_t
 646 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 647                   loff_t *ppos)
 648 {
 649         struct ftrace_event_call *call = filp->private_data;
 650         struct trace_seq *s;
 651         int r;
 652
 653         if (*ppos)
 654                 return 0;
 655
 656         s = kmalloc(sizeof(*s), GFP_KERNEL);
 657         if (!s)
 658                 return -ENOMEM;
 659
 660         trace_seq_init(s);
 661
 662         print_event_filter(call, s);
 663         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 664
 665         kfree(s);
 666
 667         return r;
 668 }
 669
 670 static ssize_t
 671 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
 672                    loff_t *ppos)
 673 {
 674         struct ftrace_event_call *call = filp->private_data;
 675         char *buf;
 676         int err;
 677
 678         if (cnt >= PAGE_SIZE)
 679                 return -EINVAL;
 680
 681         buf = (char *)__get_free_page(GFP_TEMPORARY);
 682         if (!buf)
 683                 return -ENOMEM;
 684
 685         if (copy_from_user(buf, ubuf, cnt)) {
 686                 free_page((unsigned long) buf);
 687                 return -EFAULT;
 688         }
 689         buf[cnt] = '\0';
 690
 691         err = apply_event_filter(call, buf);
 692         free_page((unsigned long) buf);
 693         if (err < 0)
 694                 return err;
 695
 696         *ppos += cnt;
 697
 698         return cnt;
 699 }
 700
 701 static ssize_t
 702 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 703                       loff_t *ppos)
 704 {
 705         struct event_subsystem *system = filp->private_data;
 706         struct trace_seq *s;
 707         int r;
 708
 709         if (*ppos)
 710                 return 0;
 711
 712         s = kmalloc(sizeof(*s), GFP_KERNEL);
 713         if (!s)
 714                 return -ENOMEM;
 715
 716         trace_seq_init(s);
 717
 718         print_subsystem_event_filter(system, s);
 719         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 720
 721         kfree(s);
 722
 723         return r;
 724 }
 725
 726 static ssize_t
 727 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
 728                        loff_t *ppos)
 729 {
 730         struct event_subsystem *system = filp->private_data;
 731         char *buf;
 732         int err;
 733
 734         if (cnt >= PAGE_SIZE)
 735                 return -EINVAL;
 736
 737         buf = (char *)__get_free_page(GFP_TEMPORARY);
 738         if (!buf)
 739                 return -ENOMEM;
 740
 741         if (copy_from_user(buf, ubuf, cnt)) {
 742                 free_page((unsigned long) buf);
 743                 return -EFAULT;
 744         }
 745         buf[cnt] = '\0';
 746
 747         err = apply_subsystem_event_filter(system, buf);
 748         free_page((unsigned long) buf);
 749         if (err < 0)
 750                 return err;
 751
 752         *ppos += cnt;
 753
 754         return cnt;
 755 }
 756
 757 static ssize_t
 758 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 759 {
 760         int (*func)(struct trace_seq *s) = filp->private_data;
 761         struct trace_seq *s;
 762         int r;
 763
 764         if (*ppos)
 765                 return 0;
 766
 767         s = kmalloc(sizeof(*s), GFP_KERNEL);
 768         if (!s)
 769                 return -ENOMEM;
 770
 771         trace_seq_init(s);
 772
 773         func(s);
 774         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 775
 776         kfree(s);
 777
 778         return r;
 779 }
 780
 781 static const struct seq_operations show_event_seq_ops = {
 782         .start = t_start,
 783         .next = t_next,
 784         .show = t_show,
 785         .stop = t_stop,
 786 };
 787
 788 static const struct seq_operations show_set_event_seq_ops = {
 789         .start = s_start,
 790         .next = s_next,
 791         .show = t_show,
 792         .stop = t_stop,
 793 };
 794
 795 static const struct file_operations ftrace_avail_fops = {
 796         .open = ftrace_event_seq_open,
 797         .read = seq_read,
 798         .llseek = seq_lseek,
 799         .release = seq_release,
 800 };
 801
 802 static const struct file_operations ftrace_set_event_fops = {
 803         .open = ftrace_event_seq_open,
 804         .read = seq_read,
 805         .write = ftrace_event_write,
 806         .llseek = seq_lseek,
 807         .release = seq_release,
 808 };
 809
 810 static const struct file_operations ftrace_enable_fops = {
 811         .open = tracing_open_generic,
 812         .read = event_enable_read,
 813         .write = event_enable_write,
 814 };
 815
 816 static const struct file_operations ftrace_event_format_fops = {
 817         .open = tracing_open_generic,
 818         .read = event_format_read,
 819 };
 820
 821 static const struct file_operations ftrace_event_id_fops = {
 822         .open = tracing_open_generic,
 823         .read = event_id_read,
 824 };
 825
 826 static const struct file_operations ftrace_event_filter_fops = {
 827         .open = tracing_open_generic,
 828         .read = event_filter_read,
 829         .write = event_filter_write,
 830 };
 831
 832 static const struct file_operations ftrace_subsystem_filter_fops = {
 833         .open = tracing_open_generic,
 834         .read = subsystem_filter_read,
 835         .write = subsystem_filter_write,
 836 };
 837
 838 static const struct file_operations ftrace_system_enable_fops = {
 839         .open = tracing_open_generic,
 840         .read = system_enable_read,
 841         .write = system_enable_write,
 842 };
 843
 844 static const struct file_operations ftrace_show_header_fops = {
 845         .open = tracing_open_generic,
 846         .read = show_header,
 847 };
 848
 849 static struct dentry *event_trace_events_dir(void)
 850 {
 851         static struct dentry *d_tracer;
 852         static struct dentry *d_events;
 853
 854         if (d_events)
 855                 return d_events;
 856
 857         d_tracer = tracing_init_dentry();
 858         if (!d_tracer)
 859                 return NULL;
 860
 861         d_events = debugfs_create_dir("events", d_tracer);
 862         if (!d_events)
 863                 pr_warning("Could not create debugfs "
 864                            "'events' directory\n");
 865
 866         return d_events;
 867 }
 868
 869 static LIST_HEAD(event_subsystems);
 870
 871 static struct dentry *
 872 event_subsystem_dir(const char *name, struct dentry *d_events)
 873 {
 874         struct event_subsystem *system;
 875         struct dentry *entry;
 876
 877         /* First see if we did not already create this dir */
 878         list_for_each_entry(system, &event_subsystems, list) {
 879                 if (strcmp(system->name, name) == 0) {
 880                         system->nr_events++;
 881                         return system->entry;
 882                 }
 883         }
 884
 885         /* need to create new entry */
 886         system = kmalloc(sizeof(*system), GFP_KERNEL);
 887         if (!system) {
 888                 pr_warning("No memory to create event subsystem %s\n",
 889                            name);
 890                 return d_events;
 891         }
 892
 893         system->entry = debugfs_create_dir(name, d_events);
 894         if (!system->entry) {
 895                 pr_warning("Could not create event subsystem %s\n",
 896                            name);
 897                 kfree(system);
 898                 return d_events;
 899         }
 900
 901         system->nr_events = 1;
 902         system->name = kstrdup(name, GFP_KERNEL);
 903         if (!system->name) {
 904                 debugfs_remove(system->entry);
 905                 kfree(system);
 906                 return d_events;
 907         }
 908
 909         list_add(&system->list, &event_subsystems);
 910
 911         system->filter = NULL;
 912
 913         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
 914         if (!system->filter) {
 915                 pr_warning("Could not allocate filter for subsystem "
 916                            "'%s'\n", name);
 917                 return system->entry;
 918         }
 919
 920         entry = debugfs_create_file("filter", 0644, system->entry, system,
 921                                     &ftrace_subsystem_filter_fops);
 922         if (!entry) {
 923                 kfree(system->filter);
 924                 system->filter = NULL;
 925                 pr_warning("Could not create debugfs "
 926                            "'%s/filter' entry\n", name);
 927         }
 928
 929         entry = trace_create_file("enable", 0644, system->entry,
 930                                   (void *)system->name,
 931                                   &ftrace_system_enable_fops);
 932
 933         return system->entry;
 934 }
 935
 936 static int
 937 event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
 938                  const struct file_operations *id,
 939                  const struct file_operations *enable,
 940                  const struct file_operations *filter,
 941                  const struct file_operations *format)
 942 {
 943         struct dentry *entry;
 944         int ret;
 945
 946         /*
 947          * If the trace point header did not define TRACE_SYSTEM
 948          * then the system would be called "TRACE_SYSTEM".
 949          */
 950         if (strcmp(call->system, TRACE_SYSTEM) != 0)
 951                 d_events = event_subsystem_dir(call->system, d_events);
 952
 953         call->dir = debugfs_create_dir(call->name, d_events);
 954         if (!call->dir) {
 955                 pr_warning("Could not create debugfs "
 956                            "'%s' directory\n", call->name);
 957                 return -1;
 958         }
 959
 960         if (call->regfunc)
 961                 entry = trace_create_file("enable", 0644, call->dir, call,
 962                                           enable);
 963
 964         if (call->id && call->profile_enable)
 965                 entry = trace_create_file("id", 0444, call->dir, call,
 966                                           id);
 967
 968         if (call->define_fields) {
 969                 ret = call->define_fields(call);
 970                 if (ret < 0) {
 971                         pr_warning("Could not initialize trace point"
 972                                    " events/%s\n", call->name);
 973                         return ret;
 974                 }
 975                 entry = trace_create_file("filter", 0644, call->dir, call,
 976                                           filter);
 977         }
 978
 979         /* A trace may not want to export its format */
 980         if (!call->show_format)
 981                 return 0;
 982
 983         entry = trace_create_file("format", 0444, call->dir, call,
 984                                   format);
 985
 986         return 0;
 987 }
 988
 989 #define for_each_event(event, start, end)                       \
 990         for (event = start;                                     \
 991              (unsigned long)event < (unsigned long)end;         \
 992              event++)
 993
 994 #ifdef CONFIG_MODULES
 995
 996 static LIST_HEAD(ftrace_module_file_list);
 997
 998 /*
 999  * Modules must own their file_operations to keep up with
1000  * reference counting.
1001  */
1002 struct ftrace_module_file_ops {
1003         struct list_head                list;
1004         struct module                   *mod;
1005         struct file_operations          id;
1006         struct file_operations          enable;
1007         struct file_operations          format;
1008         struct file_operations          filter;
1009 };
1010
1011 static void remove_subsystem_dir(const char *name)
1012 {
1013         struct event_subsystem *system;
1014
1015         if (strcmp(name, TRACE_SYSTEM) == 0)
1016                 return;
1017
1018         list_for_each_entry(system, &event_subsystems, list) {
1019                 if (strcmp(system->name, name) == 0) {
1020                         if (!--system->nr_events) {
1021                                 struct event_filter *filter = system->filter;
1022
1023                                 debugfs_remove_recursive(system->entry);
1024                                 list_del(&system->list);
1025                                 if (filter) {
1026                                         kfree(filter->filter_string);
1027                                         kfree(filter);
1028                                 }
1029                                 kfree(system->name);
1030                                 kfree(system);
1031                         }
1032                         break;
1033                 }
1034         }
1035 }
1036
1037 static struct ftrace_module_file_ops *
1038 trace_create_file_ops(struct module *mod)
1039 {
1040         struct ftrace_module_file_ops *file_ops;
1041
1042         /*
1043          * This is a bit of a PITA. To allow for correct reference
1044          * counting, modules must "own" their file_operations.
1045          * To do this, we allocate the file operations that will be
1046          * used in the event directory.
1047          */
1048
1049         file_ops = kmalloc(sizeof(*file_ops), GFP_KERNEL);
1050         if (!file_ops)
1051                 return NULL;
1052
1053         file_ops->mod = mod;
1054
1055         file_ops->id = ftrace_event_id_fops;
1056         file_ops->id.owner = mod;
1057
1058         file_ops->enable = ftrace_enable_fops;
1059         file_ops->enable.owner = mod;
1060
1061         file_ops->filter = ftrace_event_filter_fops;
1062         file_ops->filter.owner = mod;
1063
1064         file_ops->format = ftrace_event_format_fops;
1065         file_ops->format.owner = mod;
1066
1067         list_add(&file_ops->list, &ftrace_module_file_list);
1068
1069         return file_ops;
1070 }
1071
1072 static void trace_module_add_events(struct module *mod)
1073 {
1074         struct ftrace_module_file_ops *file_ops = NULL;
1075         struct ftrace_event_call *call, *start, *end;
1076         struct dentry *d_events;
1077         int ret;
1078
1079         start = mod->trace_events;
1080         end = mod->trace_events + mod->num_trace_events;
1081
1082         if (start == end)
1083                 return;
1084
1085         d_events = event_trace_events_dir();
1086         if (!d_events)
1087                 return;
1088
1089         for_each_event(call, start, end) {
1090                 /* The linker may leave blanks */
1091                 if (!call->name)
1092                         continue;
1093                 if (call->raw_init) {
1094                         ret = call->raw_init();
1095                         if (ret < 0) {
1096                                 if (ret != -ENOSYS)
1097                                         pr_warning("Could not initialize trace "
1098                                         "point events/%s\n", call->name);
1099                                 continue;
1100                         }
1101                 }
1102                 /*
1103                  * This module has events, create file ops for this module
1104                  * if not already done.
1105                  */
1106                 if (!file_ops) {
1107                         file_ops = trace_create_file_ops(mod);
1108                         if (!file_ops)
1109                                 return;
1110                 }
1111                 call->mod = mod;
1112                 list_add(&call->list, &ftrace_events);
1113                 event_create_dir(call, d_events,
1114                                  &file_ops->id, &file_ops->enable,
1115                                  &file_ops->filter, &file_ops->format);
1116         }
1117 }
1118
1119 static void trace_module_remove_events(struct module *mod)
1120 {
1121         struct ftrace_module_file_ops *file_ops;
1122         struct ftrace_event_call *call, *p;
1123         bool found = false;
1124
1125         down_write(&trace_event_mutex);
1126         list_for_each_entry_safe(call, p, &ftrace_events, list) {
1127                 if (call->mod == mod) {
1128                         found = true;
1129                         ftrace_event_enable_disable(call, 0);
1130                         if (call->event)
1131                                 __unregister_ftrace_event(call->event);
1132                         debugfs_remove_recursive(call->dir);
1133                         list_del(&call->list);
1134                         trace_destroy_fields(call);
1135                         destroy_preds(call);
1136                         remove_subsystem_dir(call->system);
1137                 }
1138         }
1139
1140         /* Now free the file_operations */
1141         list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1142                 if (file_ops->mod == mod)
1143                         break;
1144         }
1145         if (&file_ops->list != &ftrace_module_file_list) {
1146                 list_del(&file_ops->list);
1147                 kfree(file_ops);
1148         }
1149
1150         /*
1151          * It is safest to reset the ring buffer if the module being unloaded
1152          * registered any events.
1153          */
1154         if (found)
1155                 tracing_reset_current_online_cpus();
1156         up_write(&trace_event_mutex);
1157 }
1158
1159 static int trace_module_notify(struct notifier_block *self,
1160                                unsigned long val, void *data)
1161 {
1162         struct module *mod = data;
1163
1164         mutex_lock(&event_mutex);
1165         switch (val) {
1166         case MODULE_STATE_COMING:
1167                 trace_module_add_events(mod);
1168                 break;
1169         case MODULE_STATE_GOING:
1170                 trace_module_remove_events(mod);
1171                 break;
1172         }
1173         mutex_unlock(&event_mutex);
1174
1175         return 0;
1176 }
1177 #else
1178 static int trace_module_notify(struct notifier_block *self,
1179                                unsigned long val, void *data)
1180 {
1181         return 0;
1182 }
1183 #endif /* CONFIG_MODULES */
1184
1185 struct notifier_block trace_module_nb = {
1186         .notifier_call = trace_module_notify,
1187         .priority = 0,
1188 };
1189
1190 extern struct ftrace_event_call __start_ftrace_events[];
1191 extern struct ftrace_event_call __stop_ftrace_events[];
1192
1193 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
1194
1195 static __init int setup_trace_event(char *str)
1196 {
1197         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
1198         ring_buffer_expanded = 1;
1199         tracing_selftest_disabled = 1;
1200
1201         return 1;
1202 }
1203 __setup("trace_event=", setup_trace_event);
1204
1205 static __init int event_trace_init(void)
1206 {
1207         struct ftrace_event_call *call;
1208         struct dentry *d_tracer;
1209         struct dentry *entry;
1210         struct dentry *d_events;
1211         int ret;
1212         char *buf = bootup_event_buf;
1213         char *token;
1214
1215         d_tracer = tracing_init_dentry();
1216         if (!d_tracer)
1217                 return 0;
1218
1219         entry = debugfs_create_file("available_events", 0444, d_tracer,
1220                                     (void *)&show_event_seq_ops,
1221                                     &ftrace_avail_fops);
1222         if (!entry)
1223                 pr_warning("Could not create debugfs "
1224                            "'available_events' entry\n");
1225
1226         entry = debugfs_create_file("set_event", 0644, d_tracer,
1227                                     (void *)&show_set_event_seq_ops,
1228                                     &ftrace_set_event_fops);
1229         if (!entry)
1230                 pr_warning("Could not create debugfs "
1231                            "'set_event' entry\n");
1232
1233         d_events = event_trace_events_dir();
1234         if (!d_events)
1235                 return 0;
1236
1237         /* ring buffer internal formats */
1238         trace_create_file("header_page", 0444, d_events,
1239                           ring_buffer_print_page_header,
1240                           &ftrace_show_header_fops);
1241
1242         trace_create_file("header_event", 0444, d_events,
1243                           ring_buffer_print_entry_header,
1244                           &ftrace_show_header_fops);
1245
1246         trace_create_file("enable", 0644, d_events,
1247                           NULL, &ftrace_system_enable_fops);
1248
1249         for_each_event(call, __start_ftrace_events, __stop_ftrace_events) {
1250                 /* The linker may leave blanks */
1251                 if (!call->name)
1252                         continue;
1253                 if (call->raw_init) {
1254                         ret = call->raw_init();
1255                         if (ret < 0) {
1256                                 if (ret != -ENOSYS)
1257                                         pr_warning("Could not initialize trace "
1258                                         "point events/%s\n", call->name);
1259                                 continue;
1260                         }
1261                 }
1262                 list_add(&call->list, &ftrace_events);
1263                 event_create_dir(call, d_events, &ftrace_event_id_fops,
1264                                  &ftrace_enable_fops, &ftrace_event_filter_fops,
1265                                  &ftrace_event_format_fops);
1266         }
1267
1268         while (true) {
1269                 token = strsep(&buf, ",");
1270
1271                 if (!token)
1272                         break;
1273                 if (!*token)
1274                         continue;
1275
1276                 ret = ftrace_set_clr_event(token, 1);
1277                 if (ret)
1278                         pr_warning("Failed to enable trace event: %s\n", token);
1279         }
1280
1281         ret = register_module_notifier(&trace_module_nb);
1282         if (ret)
1283                 pr_warning("Failed to register trace events module notifier\n");
1284
1285         return 0;
1286 }
1287 fs_initcall(event_trace_init);
1288
1289 #ifdef CONFIG_FTRACE_STARTUP_TEST
1290
1291 static DEFINE_SPINLOCK(test_spinlock);
1292 static DEFINE_SPINLOCK(test_spinlock_irq);
1293 static DEFINE_MUTEX(test_mutex);
1294
1295 static __init void test_work(struct work_struct *dummy)
1296 {
1297         spin_lock(&test_spinlock);
1298         spin_lock_irq(&test_spinlock_irq);
1299         udelay(1);
1300         spin_unlock_irq(&test_spinlock_irq);
1301         spin_unlock(&test_spinlock);
1302
1303         mutex_lock(&test_mutex);
1304         msleep(1);
1305         mutex_unlock(&test_mutex);
1306 }
1307
1308 static __init int event_test_thread(void *unused)
1309 {
1310         void *test_malloc;
1311
1312         test_malloc = kmalloc(1234, GFP_KERNEL);
1313         if (!test_malloc)
1314                 pr_info("failed to kmalloc\n");
1315
1316         schedule_on_each_cpu(test_work);
1317
1318         kfree(test_malloc);
1319
1320         set_current_state(TASK_INTERRUPTIBLE);
1321         while (!kthread_should_stop())
1322                 schedule();
1323
1324         return 0;
1325 }
1326
1327 /*
1328  * Do various things that may trigger events.
1329  */
1330 static __init void event_test_stuff(void)
1331 {
1332         struct task_struct *test_thread;
1333
1334         test_thread = kthread_run(event_test_thread, NULL, "test-events");
1335         msleep(1);
1336         kthread_stop(test_thread);
1337 }
1338
1339 /*
1340  * For every trace event defined, we will test each trace point separately,
1341  * and then by groups, and finally all trace points.
1342  */
1343 static __init void event_trace_self_tests(void)
1344 {
1345         struct ftrace_event_call *call;
1346         struct event_subsystem *system;
1347         int ret;
1348
1349         pr_info("Running tests on trace events:\n");
1350
1351         list_for_each_entry(call, &ftrace_events, list) {
1352
1353                 /* Only test those that have a regfunc */
1354                 if (!call->regfunc)
1355                         continue;
1356
1357                 pr_info("Testing event %s: ", call->name);
1358
1359                 /*
1360                  * If an event is already enabled, someone is using
1361                  * it and the self test should not be on.
1362                  */
1363                 if (call->enabled) {
1364                         pr_warning("Enabled event during self test!\n");
1365                         WARN_ON_ONCE(1);
1366                         continue;
1367                 }
1368
1369                 ftrace_event_enable_disable(call, 1);
1370                 event_test_stuff();
1371                 ftrace_event_enable_disable(call, 0);
1372
1373                 pr_cont("OK\n");
1374         }
1375
1376         /* Now test at the sub system level */
1377
1378         pr_info("Running tests on trace event systems:\n");
1379
1380         list_for_each_entry(system, &event_subsystems, list) {
1381
1382                 /* the ftrace system is special, skip it */
1383                 if (strcmp(system->name, "ftrace") == 0)
1384                         continue;
1385
1386                 pr_info("Testing event system %s: ", system->name);
1387
1388                 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 1);
1389                 if (WARN_ON_ONCE(ret)) {
1390                         pr_warning("error enabling system %s\n",
1391                                    system->name);
1392                         continue;
1393                 }
1394
1395                 event_test_stuff();
1396
1397                 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 0);
1398                 if (WARN_ON_ONCE(ret))
1399                         pr_warning("error disabling system %s\n",
1400                                    system->name);
1401
1402                 pr_cont("OK\n");
1403         }
1404
1405         /* Test with all events enabled */
1406
1407         pr_info("Running tests on all trace events:\n");
1408         pr_info("Testing all events: ");
1409
1410         ret = __ftrace_set_clr_event(NULL, NULL, NULL, 1);
1411         if (WARN_ON_ONCE(ret)) {
1412                 pr_warning("error enabling all events\n");
1413                 return;
1414         }
1415
1416         event_test_stuff();
1417
1418         /* reset sysname */
1419         ret = __ftrace_set_clr_event(NULL, NULL, NULL, 0);
1420         if (WARN_ON_ONCE(ret)) {
1421                 pr_warning("error disabling all events\n");
1422                 return;
1423         }
1424
1425         pr_cont("OK\n");
1426 }
1427
1428 #ifdef CONFIG_FUNCTION_TRACER
1429
1430 static DEFINE_PER_CPU(atomic_t, test_event_disable);
1431
1432 static void
1433 function_test_events_call(unsigned long ip, unsigned long parent_ip)
1434 {
1435         struct ring_buffer_event *event;
1436         struct ftrace_entry *entry;
1437         unsigned long flags;
1438         long disabled;
1439         int resched;
1440         int cpu;
1441         int pc;
1442
1443         pc = preempt_count();
1444         resched = ftrace_preempt_disable();
1445         cpu = raw_smp_processor_id();
1446         disabled = atomic_inc_return(&per_cpu(test_event_disable, cpu));
1447
1448         if (disabled != 1)
1449                 goto out;
1450
1451         local_save_flags(flags);
1452
1453         event = trace_current_buffer_lock_reserve(TRACE_FN, sizeof(*entry),
1454                                                   flags, pc);
1455         if (!event)
1456                 goto out;
1457         entry   = ring_buffer_event_data(event);
1458         entry->ip                       = ip;
1459         entry->parent_ip                = parent_ip;
1460
1461         trace_nowake_buffer_unlock_commit(event, flags, pc);
1462
1463  out:
1464         atomic_dec(&per_cpu(test_event_disable, cpu));
1465         ftrace_preempt_enable(resched);
1466 }
1467
1468 static struct ftrace_ops trace_ops __initdata  =
1469 {
1470         .func = function_test_events_call,
1471 };
1472
1473 static __init void event_trace_self_test_with_function(void)
1474 {
1475         register_ftrace_function(&trace_ops);
1476         pr_info("Running tests again, along with the function tracer\n");
1477         event_trace_self_tests();
1478         unregister_ftrace_function(&trace_ops);
1479 }
1480 #else
1481 static __init void event_trace_self_test_with_function(void)
1482 {
1483 }
1484 #endif
1485
1486 static __init int event_trace_self_tests_init(void)
1487 {
1488         if (!tracing_selftest_disabled) {
1489                 event_trace_self_tests();
1490                 event_trace_self_test_with_function();
1491         }
1492
1493         return 0;
1494 }
1495
1496 late_initcall(event_trace_self_tests_init);
1497
1498 #endif