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