kernel/trace/trace_events.c

   1 /*
   2  * event tracer
   3  *
   4  * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
   5  *
   6  *  - Added format output of fields of the trace point.
   7  *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
   8  *
   9  */
  10
  11 #include <linux/workqueue.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/kthread.h>
  14 #include <linux/debugfs.h>
  15 #include <linux/uaccess.h>
  16 #include <linux/module.h>
  17 #include <linux/ctype.h>
  18 #include <linux/delay.h>
  19
  20 #include <asm/setup.h>
  21
  22 #include "trace_output.h"
  23
  24 #define TRACE_SYSTEM "TRACE_SYSTEM"
  25
  26 DEFINE_MUTEX(event_mutex);
  27
  28 LIST_HEAD(ftrace_events);
  29
  30 int trace_define_field(struct ftrace_event_call *call, const char *type,
  31                        const char *name, int offset, int size, int is_signed,
  32                        int filter_type)
  33 {
  34         struct ftrace_event_field *field;
  35
  36         field = kzalloc(sizeof(*field), GFP_KERNEL);
  37         if (!field)
  38                 goto err;
  39
  40         field->name = kstrdup(name, GFP_KERNEL);
  41         if (!field->name)
  42                 goto err;
  43
  44         field->type = kstrdup(type, GFP_KERNEL);
  45         if (!field->type)
  46                 goto err;
  47
  48         if (filter_type == FILTER_OTHER)
  49                 field->filter_type = filter_assign_type(type);
  50         else
  51                 field->filter_type = filter_type;
  52
  53         field->offset = offset;
  54         field->size = size;
  55         field->is_signed = is_signed;
  56
  57         list_add(&field->link, &call->fields);
  58
  59         return 0;
  60
  61 err:
  62         if (field) {
  63                 kfree(field->name);
  64                 kfree(field->type);
  65         }
  66         kfree(field);
  67
  68         return -ENOMEM;
  69 }
  70 EXPORT_SYMBOL_GPL(trace_define_field);
  71
  72 #define __common_field(type, item)                                      \
  73         ret = trace_define_field(call, #type, "common_" #item,          \
  74                                  offsetof(typeof(ent), item),           \
  75                                  sizeof(ent.item),                      \
  76                                  is_signed_type(type), FILTER_OTHER);   \
  77         if (ret)                                                        \
  78                 return ret;
  79
  80 int trace_define_common_fields(struct ftrace_event_call *call)
  81 {
  82         int ret;
  83         struct trace_entry ent;
  84
  85         __common_field(unsigned short, type);
  86         __common_field(unsigned char, flags);
  87         __common_field(unsigned char, preempt_count);
  88         __common_field(int, pid);
  89         __common_field(int, tgid);
  90
  91         return ret;
  92 }
  93 EXPORT_SYMBOL_GPL(trace_define_common_fields);
  94
  95 void trace_destroy_fields(struct ftrace_event_call *call)
  96 {
  97         struct ftrace_event_field *field, *next;
  98
  99         list_for_each_entry_safe(field, next, &call->fields, link) {
 100                 list_del(&field->link);
 101                 kfree(field->type);
 102                 kfree(field->name);
 103                 kfree(field);
 104         }
 105 }
 106
 107 static void ftrace_event_enable_disable(struct ftrace_event_call *call,
 108                                         int enable)
 109 {
 110         switch (enable) {
 111         case 0:
 112                 if (call->enabled) {
 113                         call->enabled = 0;
 114                         tracing_stop_cmdline_record();
 115                         call->unregfunc(call);
 116                 }
 117                 break;
 118         case 1:
 119                 if (!call->enabled) {
 120                         call->enabled = 1;
 121                         tracing_start_cmdline_record();
 122                         call->regfunc(call);
 123                 }
 124                 break;
 125         }
 126 }
 127
 128 static void ftrace_clear_events(void)
 129 {
 130         struct ftrace_event_call *call;
 131
 132         mutex_lock(&event_mutex);
 133         list_for_each_entry(call, &ftrace_events, list) {
 134                 ftrace_event_enable_disable(call, 0);
 135         }
 136         mutex_unlock(&event_mutex);
 137 }
 138
 139 /*
 140  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 141  */
 142 static int __ftrace_set_clr_event(const char *match, const char *sub,
 143                                   const char *event, int set)
 144 {
 145         struct ftrace_event_call *call;
 146         int ret = -EINVAL;
 147
 148         mutex_lock(&event_mutex);
 149         list_for_each_entry(call, &ftrace_events, list) {
 150
 151                 if (!call->name || !call->regfunc)
 152                         continue;
 153
 154                 if (match &&
 155                     strcmp(match, call->name) != 0 &&
 156                     strcmp(match, call->system) != 0)
 157                         continue;
 158
 159                 if (sub && strcmp(sub, call->system) != 0)
 160                         continue;
 161
 162                 if (event && strcmp(event, call->name) != 0)
 163                         continue;
 164
 165                 ftrace_event_enable_disable(call, set);
 166
 167                 ret = 0;
 168         }
 169         mutex_unlock(&event_mutex);
 170
 171         return ret;
 172 }
 173
 174 static int ftrace_set_clr_event(char *buf, int set)
 175 {
 176         char *event = NULL, *sub = NULL, *match;
 177
 178         /*
 179          * The buf format can be <subsystem>:<event-name>
 180          *  *:<event-name> means any event by that name.
 181          *  :<event-name> is the same.
 182          *
 183          *  <subsystem>:* means all events in that subsystem
 184          *  <subsystem>: means the same.
 185          *
 186          *  <name> (no ':') means all events in a subsystem with
 187          *  the name <name> or any event that matches <name>
 188          */
 189
 190         match = strsep(&buf, ":");
 191         if (buf) {
 192                 sub = match;
 193                 event = buf;
 194                 match = NULL;
 195
 196                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 197                         sub = NULL;
 198                 if (!strlen(event) || strcmp(event, "*") == 0)
 199                         event = NULL;
 200         }
 201
 202         return __ftrace_set_clr_event(match, sub, event, set);
 203 }
 204
 205 /**
 206  * trace_set_clr_event - enable or disable an event
 207  * @system: system name to match (NULL for any system)
 208  * @event: event name to match (NULL for all events, within system)
 209  * @set: 1 to enable, 0 to disable
 210  *
 211  * This is a way for other parts of the kernel to enable or disable
 212  * event recording.
 213  *
 214  * Returns 0 on success, -EINVAL if the parameters do not match any
 215  * registered events.
 216  */
 217 int trace_set_clr_event(const char *system, const char *event, int set)
 218 {
 219         return __ftrace_set_clr_event(NULL, system, event, set);
 220 }
 221
 222 /* 128 should be much more than enough */
 223 #define EVENT_BUF_SIZE          127
 224
 225 static ssize_t
 226 ftrace_event_write(struct file *file, const char __user *ubuf,
 227                    size_t cnt, loff_t *ppos)
 228 {
 229         size_t read = 0;
 230         int i, set = 1;
 231         ssize_t ret;
 232         char *buf;
 233         char ch;
 234
 235         if (!cnt || cnt < 0)
 236                 return 0;
 237
 238         ret = tracing_update_buffers();
 239         if (ret < 0)
 240                 return ret;
 241
 242         ret = get_user(ch, ubuf++);
 243         if (ret)
 244                 return ret;
 245         read++;
 246         cnt--;
 247
 248         /* skip white space */
 249         while (cnt && isspace(ch)) {
 250                 ret = get_user(ch, ubuf++);
 251                 if (ret)
 252                         return ret;
 253                 read++;
 254                 cnt--;
 255         }
 256
 257         /* Only white space found? */
 258         if (isspace(ch)) {
 259                 file->f_pos += read;
 260                 ret = read;
 261                 return ret;
 262         }
 263
 264         buf = kmalloc(EVENT_BUF_SIZE+1, GFP_KERNEL);
 265         if (!buf)
 266                 return -ENOMEM;
 267
 268         if (cnt > EVENT_BUF_SIZE)
 269                 cnt = EVENT_BUF_SIZE;
 270
 271         i = 0;
 272         while (cnt && !isspace(ch)) {
 273                 if (!i && ch == '!')
 274                         set = 0;
 275                 else
 276                         buf[i++] = ch;
 277
 278                 ret = get_user(ch, ubuf++);
 279                 if (ret)
 280                         goto out_free;
 281                 read++;
 282                 cnt--;
 283         }
 284         buf[i] = 0;
 285
 286         file->f_pos += read;
 287
 288         ret = ftrace_set_clr_event(buf, set);
 289         if (ret)
 290                 goto out_free;
 291
 292         ret = read;
 293
 294  out_free:
 295         kfree(buf);
 296
 297         return ret;
 298 }
 299
 300 static void *
 301 t_next(struct seq_file *m, void *v, loff_t *pos)
 302 {
 303         struct list_head *list = m->private;
 304         struct ftrace_event_call *call;
 305
 306         (*pos)++;
 307
 308         for (;;) {
 309                 if (list == &ftrace_events)
 310                         return NULL;
 311
 312                 call = list_entry(list, struct ftrace_event_call, list);
 313
 314                 /*
 315                  * The ftrace subsystem is for showing formats only.
 316                  * They can not be enabled or disabled via the event files.
 317                  */
 318                 if (call->regfunc)
 319                         break;
 320
 321                 list = list->next;
 322         }
 323
 324         m->private = list->next;
 325
 326         return call;
 327 }
 328
 329 static void *t_start(struct seq_file *m, loff_t *pos)
 330 {
 331         struct ftrace_event_call *call = NULL;
 332         loff_t l;
 333
 334         mutex_lock(&event_mutex);
 335
 336         m->private = ftrace_events.next;
 337         for (l = 0; l <= *pos; ) {
 338                 call = t_next(m, NULL, &l);
 339                 if (!call)
 340                         break;
 341         }
 342         return call;
 343 }
 344
 345 static void *
 346 s_next(struct seq_file *m, void *v, loff_t *pos)
 347 {
 348         struct list_head *list = m->private;
 349         struct ftrace_event_call *call;
 350
 351         (*pos)++;
 352
 353  retry:
 354         if (list == &ftrace_events)
 355                 return NULL;
 356
 357         call = list_entry(list, struct ftrace_event_call, list);
 358
 359         if (!call->enabled) {
 360                 list = list->next;
 361                 goto retry;
 362         }
 363
 364         m->private = list->next;
 365
 366         return call;
 367 }
 368
 369 static void *s_start(struct seq_file *m, loff_t *pos)
 370 {
 371         struct ftrace_event_call *call = NULL;
 372         loff_t l;
 373
 374         mutex_lock(&event_mutex);
 375
 376         m->private = ftrace_events.next;
 377         for (l = 0; l <= *pos; ) {
 378                 call = s_next(m, NULL, &l);
 379                 if (!call)
 380                         break;
 381         }
 382         return call;
 383 }
 384
 385 static int t_show(struct seq_file *m, void *v)
 386 {
 387         struct ftrace_event_call *call = v;
 388
 389         if (strcmp(call->system, TRACE_SYSTEM) != 0)
 390                 seq_printf(m, "%s:", call->system);
 391         seq_printf(m, "%s\n", call->name);
 392
 393         return 0;
 394 }
 395
 396 static void t_stop(struct seq_file *m, void *p)
 397 {
 398         mutex_unlock(&event_mutex);
 399 }
 400
 401 static int
 402 ftrace_event_seq_open(struct inode *inode, struct file *file)
 403 {
 404         const struct seq_operations *seq_ops;
 405
 406         if ((file->f_mode & FMODE_WRITE) &&
 407             (file->f_flags & O_TRUNC))
 408                 ftrace_clear_events();
 409
 410         seq_ops = inode->i_private;
 411         return seq_open(file, seq_ops);
 412 }
 413
 414 static ssize_t
 415 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 416                   loff_t *ppos)
 417 {
 418         struct ftrace_event_call *call = filp->private_data;
 419         char *buf;
 420
 421         if (call->enabled)
 422                 buf = "1\n";
 423         else
 424                 buf = "0\n";
 425
 426         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 427 }
 428
 429 static ssize_t
 430 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 431                    loff_t *ppos)
 432 {
 433         struct ftrace_event_call *call = filp->private_data;
 434         char buf[64];
 435         unsigned long val;
 436         int ret;
 437
 438         if (cnt >= sizeof(buf))
 439                 return -EINVAL;
 440
 441         if (copy_from_user(&buf, ubuf, cnt))
 442                 return -EFAULT;
 443
 444         buf[cnt] = 0;
 445
 446         ret = strict_strtoul(buf, 10, &val);
 447         if (ret < 0)
 448                 return ret;
 449
 450         ret = tracing_update_buffers();
 451         if (ret < 0)
 452                 return ret;
 453
 454         switch (val) {
 455         case 0:
 456         case 1:
 457                 mutex_lock(&event_mutex);
 458                 ftrace_event_enable_disable(call, val);
 459                 mutex_unlock(&event_mutex);
 460                 break;
 461
 462         default:
 463                 return -EINVAL;
 464         }
 465
 466         *ppos += cnt;
 467
 468         return cnt;
 469 }
 470
 471 static ssize_t
 472 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 473                    loff_t *ppos)
 474 {
 475         const char set_to_char[4] = { '?', '0', '1', 'X' };
 476         const char *system = filp->private_data;
 477         struct ftrace_event_call *call;
 478         char buf[2];
 479         int set = 0;
 480         int ret;
 481
 482         mutex_lock(&event_mutex);
 483         list_for_each_entry(call, &ftrace_events, list) {
 484                 if (!call->name || !call->regfunc)
 485                         continue;
 486
 487                 if (system && strcmp(call->system, system) != 0)
 488                         continue;
 489
 490                 /*
 491                  * We need to find out if all the events are set
 492                  * or if all events or cleared, or if we have
 493                  * a mixture.
 494                  */
 495                 set |= (1 << !!call->enabled);
 496
 497                 /*
 498                  * If we have a mixture, no need to look further.
 499                  */
 500                 if (set == 3)
 501                         break;
 502         }
 503         mutex_unlock(&event_mutex);
 504
 505         buf[0] = set_to_char[set];
 506         buf[1] = '\n';
 507
 508         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 509
 510         return ret;
 511 }
 512
 513 static ssize_t
 514 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 515                     loff_t *ppos)
 516 {
 517         const char *system = filp->private_data;
 518         unsigned long val;
 519         char buf[64];
 520         ssize_t ret;
 521
 522         if (cnt >= sizeof(buf))
 523                 return -EINVAL;
 524
 525         if (copy_from_user(&buf, ubuf, cnt))
 526                 return -EFAULT;
 527
 528         buf[cnt] = 0;
 529
 530         ret = strict_strtoul(buf, 10, &val);
 531         if (ret < 0)
 532                 return ret;
 533
 534         ret = tracing_update_buffers();
 535         if (ret < 0)
 536                 return ret;
 537
 538         if (val != 0 && val != 1)
 539                 return -EINVAL;
 540
 541         ret = __ftrace_set_clr_event(NULL, system, NULL, val);
 542         if (ret)
 543                 goto out;
 544
 545         ret = cnt;
 546
 547 out:
 548         *ppos += cnt;
 549
 550         return ret;
 551 }
 552
 553 extern char *__bad_type_size(void);
 554
 555 #undef FIELD
 556 #define FIELD(type, name)                                               \
 557         sizeof(type) != sizeof(field.name) ? __bad_type_size() :        \
 558         #type, "common_" #name, offsetof(typeof(field), name),          \
 559                 sizeof(field.name)
 560
 561 static int trace_write_header(struct trace_seq *s)
 562 {
 563         struct trace_entry field;
 564
 565         /* struct trace_entry */
 566         return trace_seq_printf(s,
 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                                 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
 572                                 "\n",
 573                                 FIELD(unsigned short, type),
 574                                 FIELD(unsigned char, flags),
 575                                 FIELD(unsigned char, preempt_count),
 576                                 FIELD(int, pid),
 577                                 FIELD(int, tgid));
 578 }
 579
 580 static ssize_t
 581 event_format_read(struct file *filp, char __user *ubuf, size_t cnt,
 582                   loff_t *ppos)
 583 {
 584         struct ftrace_event_call *call = filp->private_data;
 585         struct trace_seq *s;
 586         char *buf;
 587         int r;
 588
 589         if (*ppos)
 590                 return 0;
 591
 592         s = kmalloc(sizeof(*s), GFP_KERNEL);
 593         if (!s)
 594                 return -ENOMEM;
 595
 596         trace_seq_init(s);
 597
 598         /* If any of the first writes fail, so will the show_format. */
 599
 600         trace_seq_printf(s, "name: %s\n", call->name);
 601         trace_seq_printf(s, "ID: %d\n", call->id);
 602         trace_seq_printf(s, "format:\n");
 603         trace_write_header(s);
 604
 605         r = call->show_format(call, s);
 606         if (!r) {
 607                 /*
 608                  * ug!  The format output is bigger than a PAGE!!
 609                  */
 610                 buf = "FORMAT TOO BIG\n";
 611                 r = simple_read_from_buffer(ubuf, cnt, ppos,
 612                                               buf, strlen(buf));
 613                 goto out;
 614         }
 615
 616         r = simple_read_from_buffer(ubuf, cnt, ppos,
 617                                     s->buffer, s->len);
 618  out:
 619         kfree(s);
 620         return r;
 621 }
 622
 623 static ssize_t
 624 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 625 {
 626         struct ftrace_event_call *call = filp->private_data;
 627         struct trace_seq *s;
 628         int r;
 629
 630         if (*ppos)
 631                 return 0;
 632
 633         s = kmalloc(sizeof(*s), GFP_KERNEL);
 634         if (!s)
 635                 return -ENOMEM;
 636
 637         trace_seq_init(s);
 638         trace_seq_printf(s, "%d\n", call->id);
 639
 640         r = simple_read_from_buffer(ubuf, cnt, ppos,
 641                                     s->buffer, s->len);
 642         kfree(s);
 643         return r;
 644 }
 645
 646 static ssize_t
 647 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 648                   loff_t *ppos)
 649 {
 650         struct ftrace_event_call *call = filp->private_data;
 651         struct trace_seq *s;
 652         int r;
 653
 654         if (*ppos)
 655                 return 0;
 656
 657         s = kmalloc(sizeof(*s), GFP_KERNEL);
 658         if (!s)
 659                 return -ENOMEM;
 660
 661         trace_seq_init(s);
 662
 663         print_event_filter(call, s);
 664         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 665
 666         kfree(s);
 667
 668         return r;
 669 }
 670
 671 static ssize_t
 672 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
 673                    loff_t *ppos)
 674 {
 675         struct ftrace_event_call *call = filp->private_data;
 676         char *buf;
 677         int err;
 678
 679         if (cnt >= PAGE_SIZE)
 680                 return -EINVAL;
 681
 682         buf = (char *)__get_free_page(GFP_TEMPORARY);
 683         if (!buf)
 684                 return -ENOMEM;
 685
 686         if (copy_from_user(buf, ubuf, cnt)) {
 687                 free_page((unsigned long) buf);
 688                 return -EFAULT;
 689         }
 690         buf[cnt] = '\0';
 691
 692         err = apply_event_filter(call, buf);
 693         free_page((unsigned long) buf);
 694         if (err < 0)
 695                 return err;
 696
 697         *ppos += cnt;
 698
 699         return cnt;
 700 }
 701
 702 static ssize_t
 703 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 704                       loff_t *ppos)
 705 {
 706         struct event_subsystem *system = filp->private_data;
 707         struct trace_seq *s;
 708         int r;
 709
 710         if (*ppos)
 711                 return 0;
 712
 713         s = kmalloc(sizeof(*s), GFP_KERNEL);
 714         if (!s)
 715                 return -ENOMEM;
 716
 717         trace_seq_init(s);
 718
 719         print_subsystem_event_filter(system, s);
 720         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 721
 722         kfree(s);
 723
 724         return r;
 725 }
 726
 727 static ssize_t
 728 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
 729                        loff_t *ppos)
 730 {
 731         struct event_subsystem *system = filp->private_data;
 732         char *buf;
 733         int err;
 734
 735         if (cnt >= PAGE_SIZE)
 736                 return -EINVAL;
 737
 738         buf = (char *)__get_free_page(GFP_TEMPORARY);
 739         if (!buf)
 740                 return -ENOMEM;
 741
 742         if (copy_from_user(buf, ubuf, cnt)) {
 743                 free_page((unsigned long) buf);
 744                 return -EFAULT;
 745         }
 746         buf[cnt] = '\0';
 747
 748         err = apply_subsystem_event_filter(system, buf);
 749         free_page((unsigned long) buf);
 750         if (err < 0)
 751                 return err;
 752
 753         *ppos += cnt;
 754
 755         return cnt;
 756 }
 757
 758 static ssize_t
 759 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 760 {
 761         int (*func)(struct trace_seq *s) = filp->private_data;
 762         struct trace_seq *s;
 763         int r;
 764
 765         if (*ppos)
 766                 return 0;
 767
 768         s = kmalloc(sizeof(*s), GFP_KERNEL);
 769         if (!s)
 770                 return -ENOMEM;
 771
 772         trace_seq_init(s);
 773
 774         func(s);
 775         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 776
 777         kfree(s);
 778
 779         return r;
 780 }
 781
 782 static const struct seq_operations show_event_seq_ops = {
 783         .start = t_start,
 784         .next = t_next,
 785         .show = t_show,
 786         .stop = t_stop,
 787 };
 788
 789 static const struct seq_operations show_set_event_seq_ops = {
 790         .start = s_start,
 791         .next = s_next,
 792         .show = t_show,
 793         .stop = t_stop,
 794 };
 795
 796 static const struct file_operations ftrace_avail_fops = {
 797         .open = ftrace_event_seq_open,
 798         .read = seq_read,
 799         .llseek = seq_lseek,
 800         .release = seq_release,
 801 };
 802
 803 static const struct file_operations ftrace_set_event_fops = {
 804         .open = ftrace_event_seq_open,
 805         .read = seq_read,
 806         .write = ftrace_event_write,
 807         .llseek = seq_lseek,
 808         .release = seq_release,
 809 };
 810
 811 static const struct file_operations ftrace_enable_fops = {
 812         .open = tracing_open_generic,
 813         .read = event_enable_read,
 814         .write = event_enable_write,
 815 };
 816
 817 static const struct file_operations ftrace_event_format_fops = {
 818         .open = tracing_open_generic,
 819         .read = event_format_read,
 820 };
 821
 822 static const struct file_operations ftrace_event_id_fops = {
 823         .open = tracing_open_generic,
 824         .read = event_id_read,
 825 };
 826
 827 static const struct file_operations ftrace_event_filter_fops = {
 828         .open = tracing_open_generic,
 829         .read = event_filter_read,
 830         .write = event_filter_write,
 831 };
 832
 833 static const struct file_operations ftrace_subsystem_filter_fops = {
 834         .open = tracing_open_generic,
 835         .read = subsystem_filter_read,
 836         .write = subsystem_filter_write,
 837 };
 838
 839 static const struct file_operations ftrace_system_enable_fops = {
 840         .open = tracing_open_generic,
 841         .read = system_enable_read,
 842         .write = system_enable_write,
 843 };
 844
 845 static const struct file_operations ftrace_show_header_fops = {
 846         .open = tracing_open_generic,
 847         .read = show_header,
 848 };
 849
 850 static struct dentry *event_trace_events_dir(void)
 851 {
 852         static struct dentry *d_tracer;
 853         static struct dentry *d_events;
 854
 855         if (d_events)
 856                 return d_events;
 857
 858         d_tracer = tracing_init_dentry();
 859         if (!d_tracer)
 860                 return NULL;
 861
 862         d_events = debugfs_create_dir("events", d_tracer);
 863         if (!d_events)
 864                 pr_warning("Could not create debugfs "
 865                            "'events' directory\n");
 866
 867         return d_events;
 868 }
 869
 870 static LIST_HEAD(event_subsystems);
 871
 872 static struct dentry *
 873 event_subsystem_dir(const char *name, struct dentry *d_events)
 874 {
 875         struct event_subsystem *system;
 876         struct dentry *entry;
 877
 878         /* First see if we did not already create this dir */
 879         list_for_each_entry(system, &event_subsystems, list) {
 880                 if (strcmp(system->name, name) == 0) {
 881                         system->nr_events++;
 882                         return system->entry;
 883                 }
 884         }
 885
 886         /* need to create new entry */
 887         system = kmalloc(sizeof(*system), GFP_KERNEL);
 888         if (!system) {
 889                 pr_warning("No memory to create event subsystem %s\n",
 890                            name);
 891                 return d_events;
 892         }
 893
 894         system->entry = debugfs_create_dir(name, d_events);
 895         if (!system->entry) {
 896                 pr_warning("Could not create event subsystem %s\n",
 897                            name);
 898                 kfree(system);
 899                 return d_events;
 900         }
 901
 902         system->nr_events = 1;
 903         system->name = kstrdup(name, GFP_KERNEL);
 904         if (!system->name) {
 905                 debugfs_remove(system->entry);
 906                 kfree(system);
 907                 return d_events;
 908         }
 909
 910         list_add(&system->list, &event_subsystems);
 911
 912         system->filter = NULL;
 913
 914         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
 915         if (!system->filter) {
 916                 pr_warning("Could not allocate filter for subsystem "
 917                            "'%s'\n", name);
 918                 return system->entry;
 919         }
 920
 921         entry = debugfs_create_file("filter", 0644, system->entry, system,
 922                                     &ftrace_subsystem_filter_fops);
 923         if (!entry) {
 924                 kfree(system->filter);
 925                 system->filter = NULL;
 926                 pr_warning("Could not create debugfs "
 927                            "'%s/filter' entry\n", name);
 928         }
 929
 930         entry = trace_create_file("enable", 0644, system->entry,
 931                                   (void *)system->name,
 932                                   &ftrace_system_enable_fops);
 933
 934         return system->entry;
 935 }
 936
 937 static int
 938 event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
 939                  const struct file_operations *id,
 940                  const struct file_operations *enable,
 941                  const struct file_operations *filter,
 942                  const struct file_operations *format)
 943 {
 944         struct dentry *entry;
 945         int ret;
 946
 947         /*
 948          * If the trace point header did not define TRACE_SYSTEM
 949          * then the system would be called "TRACE_SYSTEM".
 950          */
 951         if (strcmp(call->system, TRACE_SYSTEM) != 0)
 952                 d_events = event_subsystem_dir(call->system, d_events);
 953
 954         call->dir = debugfs_create_dir(call->name, d_events);
 955         if (!call->dir) {
 956                 pr_warning("Could not create debugfs "
 957                            "'%s' directory\n", call->name);
 958                 return -1;
 959         }
 960
 961         if (call->regfunc)
 962                 entry = trace_create_file("enable", 0644, call->dir, call,
 963                                           enable);
 964
 965         if (call->id && call->profile_enable)
 966                 entry = trace_create_file("id", 0444, call->dir, call,
 967                                           id);
 968
 969         if (call->define_fields) {
 970                 ret = call->define_fields(call);
 971                 if (ret < 0) {
 972                         pr_warning("Could not initialize trace point"
 973                                    " events/%s\n", call->name);
 974                         return ret;
 975                 }
 976                 entry = trace_create_file("filter", 0644, call->dir, call,
 977                                           filter);
 978         }
 979
 980         /* A trace may not want to export its format */
 981         if (!call->show_format)
 982                 return 0;
 983
 984         entry = trace_create_file("format", 0444, call->dir, call,
 985                                   format);
 986
 987         return 0;
 988 }
 989
 990 static int __trace_add_event_call(struct ftrace_event_call *call)
 991 {
 992         struct dentry *d_events;
 993         int ret;
 994
 995         if (!call->name)
 996                 return -EINVAL;
 997
 998         if (call->raw_init) {
 999                 ret = call->raw_init(call);
1000                 if (ret < 0) {
1001                         if (ret != -ENOSYS)
1002                                 pr_warning("Could not initialize trace "
1003                                 "events/%s\n", call->name);
1004                         return ret;
1005                 }
1006         }
1007
1008         d_events = event_trace_events_dir();
1009         if (!d_events)
1010                 return -ENOENT;
1011
1012         list_add(&call->list, &ftrace_events);
1013         return event_create_dir(call, d_events, &ftrace_event_id_fops,
1014                                 &ftrace_enable_fops, &ftrace_event_filter_fops,
1015                                 &ftrace_event_format_fops);
1016 }
1017
1018 /* Add an additional event_call dynamically */
1019 int trace_add_event_call(struct ftrace_event_call *call)
1020 {
1021         int ret;
1022         mutex_lock(&event_mutex);
1023         ret = __trace_add_event_call(call);
1024         mutex_unlock(&event_mutex);
1025         return ret;
1026 }
1027
1028 static void remove_subsystem_dir(const char *name)
1029 {
1030         struct event_subsystem *system;
1031
1032         if (strcmp(name, TRACE_SYSTEM) == 0)
1033                 return;
1034
1035         list_for_each_entry(system, &event_subsystems, list) {
1036                 if (strcmp(system->name, name) == 0) {
1037                         if (!--system->nr_events) {
1038                                 struct event_filter *filter = system->filter;
1039
1040                                 debugfs_remove_recursive(system->entry);
1041                                 list_del(&system->list);
1042                                 if (filter) {
1043                                         kfree(filter->filter_string);
1044                                         kfree(filter);
1045                                 }
1046                                 kfree(system->name);
1047                                 kfree(system);
1048                         }
1049                         break;
1050                 }
1051         }
1052 }
1053
1054 static void __trace_remove_event_call(struct ftrace_event_call *call)
1055 {
1056         ftrace_event_enable_disable(call, 0);
1057         if (call->event)
1058                 __unregister_ftrace_event(call->event);
1059         debugfs_remove_recursive(call->dir);
1060         list_del(&call->list);
1061         trace_destroy_fields(call);
1062         destroy_preds(call);
1063         remove_subsystem_dir(call->system);
1064 }
1065
1066 /* Remove an event_call */
1067 void trace_remove_event_call(struct ftrace_event_call *call)
1068 {
1069         mutex_lock(&event_mutex);
1070         __trace_remove_event_call(call);
1071         mutex_unlock(&event_mutex);
1072 }
1073
1074 #define for_each_event(event, start, end)                       \
1075         for (event = start;                                     \
1076              (unsigned long)event < (unsigned long)end;         \
1077              event++)
1078
1079 #ifdef CONFIG_MODULES
1080
1081 static LIST_HEAD(ftrace_module_file_list);
1082
1083 /*
1084  * Modules must own their file_operations to keep up with
1085  * reference counting.
1086  */
1087 struct ftrace_module_file_ops {
1088         struct list_head                list;
1089         struct module                   *mod;
1090         struct file_operations          id;
1091         struct file_operations          enable;
1092         struct file_operations          format;
1093         struct file_operations          filter;
1094 };
1095
1096 static struct ftrace_module_file_ops *
1097 trace_create_file_ops(struct module *mod)
1098 {
1099         struct ftrace_module_file_ops *file_ops;
1100
1101         /*
1102          * This is a bit of a PITA. To allow for correct reference
1103          * counting, modules must "own" their file_operations.
1104          * To do this, we allocate the file operations that will be
1105          * used in the event directory.
1106          */
1107
1108         file_ops = kmalloc(sizeof(*file_ops), GFP_KERNEL);
1109         if (!file_ops)
1110                 return NULL;
1111
1112         file_ops->mod = mod;
1113
1114         file_ops->id = ftrace_event_id_fops;
1115         file_ops->id.owner = mod;
1116
1117         file_ops->enable = ftrace_enable_fops;
1118         file_ops->enable.owner = mod;
1119
1120         file_ops->filter = ftrace_event_filter_fops;
1121         file_ops->filter.owner = mod;
1122
1123         file_ops->format = ftrace_event_format_fops;
1124         file_ops->format.owner = mod;
1125
1126         list_add(&file_ops->list, &ftrace_module_file_list);
1127
1128         return file_ops;
1129 }
1130
1131 static void trace_module_add_events(struct module *mod)
1132 {
1133         struct ftrace_module_file_ops *file_ops = NULL;
1134         struct ftrace_event_call *call, *start, *end;
1135         struct dentry *d_events;
1136         int ret;
1137
1138         start = mod->trace_events;
1139         end = mod->trace_events + mod->num_trace_events;
1140
1141         if (start == end)
1142                 return;
1143
1144         d_events = event_trace_events_dir();
1145         if (!d_events)
1146                 return;
1147
1148         for_each_event(call, start, end) {
1149                 /* The linker may leave blanks */
1150                 if (!call->name)
1151                         continue;
1152                 if (call->raw_init) {
1153                         ret = call->raw_init(call);
1154                         if (ret < 0) {
1155                                 if (ret != -ENOSYS)
1156                                         pr_warning("Could not initialize trace "
1157                                         "point events/%s\n", call->name);
1158                                 continue;
1159                         }
1160                 }
1161                 /*
1162                  * This module has events, create file ops for this module
1163                  * if not already done.
1164                  */
1165                 if (!file_ops) {
1166                         file_ops = trace_create_file_ops(mod);
1167                         if (!file_ops)
1168                                 return;
1169                 }
1170                 call->mod = mod;
1171                 list_add(&call->list, &ftrace_events);
1172                 event_create_dir(call, d_events,
1173                                  &file_ops->id, &file_ops->enable,
1174                                  &file_ops->filter, &file_ops->format);
1175         }
1176 }
1177
1178 static void trace_module_remove_events(struct module *mod)
1179 {
1180         struct ftrace_module_file_ops *file_ops;
1181         struct ftrace_event_call *call, *p;
1182         bool found = false;
1183
1184         down_write(&trace_event_mutex);
1185         list_for_each_entry_safe(call, p, &ftrace_events, list) {
1186                 if (call->mod == mod) {
1187                         found = true;
1188                         __trace_remove_event_call(call);
1189                 }
1190         }
1191
1192         /* Now free the file_operations */
1193         list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1194                 if (file_ops->mod == mod)
1195                         break;
1196         }
1197         if (&file_ops->list != &ftrace_module_file_list) {
1198                 list_del(&file_ops->list);
1199                 kfree(file_ops);
1200         }
1201
1202         /*
1203          * It is safest to reset the ring buffer if the module being unloaded
1204          * registered any events.
1205          */
1206         if (found)
1207                 tracing_reset_current_online_cpus();
1208         up_write(&trace_event_mutex);
1209 }
1210
1211 static int trace_module_notify(struct notifier_block *self,
1212                                unsigned long val, void *data)
1213 {
1214         struct module *mod = data;
1215
1216         mutex_lock(&event_mutex);
1217         switch (val) {
1218         case MODULE_STATE_COMING:
1219                 trace_module_add_events(mod);
1220                 break;
1221         case MODULE_STATE_GOING:
1222                 trace_module_remove_events(mod);
1223                 break;
1224         }
1225         mutex_unlock(&event_mutex);
1226
1227         return 0;
1228 }
1229 #else
1230 static int trace_module_notify(struct notifier_block *self,
1231                                unsigned long val, void *data)
1232 {
1233         return 0;
1234 }
1235 #endif /* CONFIG_MODULES */
1236
1237 struct notifier_block trace_module_nb = {
1238         .notifier_call = trace_module_notify,
1239         .priority = 0,
1240 };
1241
1242 extern struct ftrace_event_call __start_ftrace_events[];
1243 extern struct ftrace_event_call __stop_ftrace_events[];
1244
1245 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
1246
1247 static __init int setup_trace_event(char *str)
1248 {
1249         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
1250         ring_buffer_expanded = 1;
1251         tracing_selftest_disabled = 1;
1252
1253         return 1;
1254 }
1255 __setup("trace_event=", setup_trace_event);
1256
1257 static __init int event_trace_init(void)
1258 {
1259         struct ftrace_event_call *call;
1260         struct dentry *d_tracer;
1261         struct dentry *entry;
1262         struct dentry *d_events;
1263         int ret;
1264         char *buf = bootup_event_buf;
1265         char *token;
1266
1267         d_tracer = tracing_init_dentry();
1268         if (!d_tracer)
1269                 return 0;
1270
1271         entry = debugfs_create_file("available_events", 0444, d_tracer,
1272                                     (void *)&show_event_seq_ops,
1273                                     &ftrace_avail_fops);
1274         if (!entry)
1275                 pr_warning("Could not create debugfs "
1276                            "'available_events' entry\n");
1277
1278         entry = debugfs_create_file("set_event", 0644, d_tracer,
1279                                     (void *)&show_set_event_seq_ops,
1280                                     &ftrace_set_event_fops);
1281         if (!entry)
1282                 pr_warning("Could not create debugfs "
1283                            "'set_event' entry\n");
1284
1285         d_events = event_trace_events_dir();
1286         if (!d_events)
1287                 return 0;
1288
1289         /* ring buffer internal formats */
1290         trace_create_file("header_page", 0444, d_events,
1291                           ring_buffer_print_page_header,
1292                           &ftrace_show_header_fops);
1293
1294         trace_create_file("header_event", 0444, d_events,
1295                           ring_buffer_print_entry_header,
1296                           &ftrace_show_header_fops);
1297
1298         trace_create_file("enable", 0644, d_events,
1299                           NULL, &ftrace_system_enable_fops);
1300
1301         for_each_event(call, __start_ftrace_events, __stop_ftrace_events) {
1302                 /* The linker may leave blanks */
1303                 if (!call->name)
1304                         continue;
1305                 if (call->raw_init) {
1306                         ret = call->raw_init(call);
1307                         if (ret < 0) {
1308                                 if (ret != -ENOSYS)
1309                                         pr_warning("Could not initialize trace "
1310                                         "point events/%s\n", call->name);
1311                                 continue;
1312                         }
1313                 }
1314                 list_add(&call->list, &ftrace_events);
1315                 event_create_dir(call, d_events, &ftrace_event_id_fops,
1316                                  &ftrace_enable_fops, &ftrace_event_filter_fops,
1317                                  &ftrace_event_format_fops);
1318         }
1319
1320         while (true) {
1321                 token = strsep(&buf, ",");
1322
1323                 if (!token)
1324                         break;
1325                 if (!*token)
1326                         continue;
1327
1328                 ret = ftrace_set_clr_event(token, 1);
1329                 if (ret)
1330                         pr_warning("Failed to enable trace event: %s\n", token);
1331         }
1332
1333         ret = register_module_notifier(&trace_module_nb);
1334         if (ret)
1335                 pr_warning("Failed to register trace events module notifier\n");
1336
1337         return 0;
1338 }
1339 fs_initcall(event_trace_init);
1340
1341 #ifdef CONFIG_FTRACE_STARTUP_TEST
1342
1343 static DEFINE_SPINLOCK(test_spinlock);
1344 static DEFINE_SPINLOCK(test_spinlock_irq);
1345 static DEFINE_MUTEX(test_mutex);
1346
1347 static __init void test_work(struct work_struct *dummy)
1348 {
1349         spin_lock(&test_spinlock);
1350         spin_lock_irq(&test_spinlock_irq);
1351         udelay(1);
1352         spin_unlock_irq(&test_spinlock_irq);
1353         spin_unlock(&test_spinlock);
1354
1355         mutex_lock(&test_mutex);
1356         msleep(1);
1357         mutex_unlock(&test_mutex);
1358 }
1359
1360 static __init int event_test_thread(void *unused)
1361 {
1362         void *test_malloc;
1363
1364         test_malloc = kmalloc(1234, GFP_KERNEL);
1365         if (!test_malloc)
1366                 pr_info("failed to kmalloc\n");
1367
1368         schedule_on_each_cpu(test_work);
1369
1370         kfree(test_malloc);
1371
1372         set_current_state(TASK_INTERRUPTIBLE);
1373         while (!kthread_should_stop())
1374                 schedule();
1375
1376         return 0;
1377 }
1378
1379 /*
1380  * Do various things that may trigger events.
1381  */
1382 static __init void event_test_stuff(void)
1383 {
1384         struct task_struct *test_thread;
1385
1386         test_thread = kthread_run(event_test_thread, NULL, "test-events");
1387         msleep(1);
1388         kthread_stop(test_thread);
1389 }
1390
1391 /*
1392  * For every trace event defined, we will test each trace point separately,
1393  * and then by groups, and finally all trace points.
1394  */
1395 static __init void event_trace_self_tests(void)
1396 {
1397         struct ftrace_event_call *call;
1398         struct event_subsystem *system;
1399         int ret;
1400
1401         pr_info("Running tests on trace events:\n");
1402
1403         list_for_each_entry(call, &ftrace_events, list) {
1404
1405                 /* Only test those that have a regfunc */
1406                 if (!call->regfunc)
1407                         continue;
1408
1409                 pr_info("Testing event %s: ", call->name);
1410
1411                 /*
1412                  * If an event is already enabled, someone is using
1413                  * it and the self test should not be on.
1414                  */
1415                 if (call->enabled) {
1416                         pr_warning("Enabled event during self test!\n");
1417                         WARN_ON_ONCE(1);
1418                         continue;
1419                 }
1420
1421                 ftrace_event_enable_disable(call, 1);
1422                 event_test_stuff();
1423                 ftrace_event_enable_disable(call, 0);
1424
1425                 pr_cont("OK\n");
1426         }
1427
1428         /* Now test at the sub system level */
1429
1430         pr_info("Running tests on trace event systems:\n");
1431
1432         list_for_each_entry(system, &event_subsystems, list) {
1433
1434                 /* the ftrace system is special, skip it */
1435                 if (strcmp(system->name, "ftrace") == 0)
1436                         continue;
1437
1438                 pr_info("Testing event system %s: ", system->name);
1439
1440                 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 1);
1441                 if (WARN_ON_ONCE(ret)) {
1442                         pr_warning("error enabling system %s\n",
1443                                    system->name);
1444                         continue;
1445                 }
1446
1447                 event_test_stuff();
1448
1449                 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 0);
1450                 if (WARN_ON_ONCE(ret))
1451                         pr_warning("error disabling system %s\n",
1452                                    system->name);
1453
1454                 pr_cont("OK\n");
1455         }
1456
1457         /* Test with all events enabled */
1458
1459         pr_info("Running tests on all trace events:\n");
1460         pr_info("Testing all events: ");
1461
1462         ret = __ftrace_set_clr_event(NULL, NULL, NULL, 1);
1463         if (WARN_ON_ONCE(ret)) {
1464                 pr_warning("error enabling all events\n");
1465                 return;
1466         }
1467
1468         event_test_stuff();
1469
1470         /* reset sysname */
1471         ret = __ftrace_set_clr_event(NULL, NULL, NULL, 0);
1472         if (WARN_ON_ONCE(ret)) {
1473                 pr_warning("error disabling all events\n");
1474                 return;
1475         }
1476
1477         pr_cont("OK\n");
1478 }
1479
1480 #ifdef CONFIG_FUNCTION_TRACER
1481
1482 static DEFINE_PER_CPU(atomic_t, test_event_disable);
1483
1484 static void
1485 function_test_events_call(unsigned long ip, unsigned long parent_ip)
1486 {
1487         struct ring_buffer_event *event;
1488         struct ring_buffer *buffer;
1489         struct ftrace_entry *entry;
1490         unsigned long flags;
1491         long disabled;
1492         int resched;
1493         int cpu;
1494         int pc;
1495
1496         pc = preempt_count();
1497         resched = ftrace_preempt_disable();
1498         cpu = raw_smp_processor_id();
1499         disabled = atomic_inc_return(&per_cpu(test_event_disable, cpu));
1500
1501         if (disabled != 1)
1502                 goto out;
1503
1504         local_save_flags(flags);
1505
1506         event = trace_current_buffer_lock_reserve(&buffer,
1507                                                   TRACE_FN, sizeof(*entry),
1508                                                   flags, pc);
1509         if (!event)
1510                 goto out;
1511         entry   = ring_buffer_event_data(event);
1512         entry->ip                       = ip;
1513         entry->parent_ip                = parent_ip;
1514
1515         trace_nowake_buffer_unlock_commit(buffer, event, flags, pc);
1516
1517  out:
1518         atomic_dec(&per_cpu(test_event_disable, cpu));
1519         ftrace_preempt_enable(resched);
1520 }
1521
1522 static struct ftrace_ops trace_ops __initdata  =
1523 {
1524         .func = function_test_events_call,
1525 };
1526
1527 static __init void event_trace_self_test_with_function(void)
1528 {
1529         register_ftrace_function(&trace_ops);
1530         pr_info("Running tests again, along with the function tracer\n");
1531         event_trace_self_tests();
1532         unregister_ftrace_function(&trace_ops);
1533 }
1534 #else
1535 static __init void event_trace_self_test_with_function(void)
1536 {
1537 }
1538 #endif
1539
1540 static __init int event_trace_self_tests_init(void)
1541 {
1542         if (!tracing_selftest_disabled) {
1543                 event_trace_self_tests();
1544                 event_trace_self_test_with_function();
1545         }
1546
1547         return 0;
1548 }
1549
1550 late_initcall(event_trace_self_tests_init);
1551
1552 #endif