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

   1 /*
   2  * event tracer
   3  *
   4  * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
   5  *
   6  *  - Added format output of fields of the trace point.
   7  *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
   8  *
   9  */
  10
  11 #include <linux/workqueue.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/kthread.h>
  14 #include <linux/debugfs.h>
  15 #include <linux/uaccess.h>
  16 #include <linux/module.h>
  17 #include <linux/ctype.h>
  18 #include <linux/delay.h>
  19
  20 #include <asm/setup.h>
  21
  22 #include "trace_output.h"
  23
  24 #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 extern char *__bad_type_size(void);
 524
 525 #undef FIELD
 526 #define FIELD(type, name)                                               \
 527         sizeof(type) != sizeof(field.name) ? __bad_type_size() :        \
 528         #type, "common_" #name, offsetof(typeof(field), name),          \
 529                 sizeof(field.name), is_signed_type(type)
 530
 531 static ssize_t
 532 event_format_read(struct file *filp, char __user *ubuf, size_t cnt,
 533                   loff_t *ppos)
 534 {
 535         struct ftrace_event_call *call = filp->private_data;
 536         struct ftrace_event_field *field;
 537         struct trace_seq *s;
 538         int common_field_count = 5;
 539         char *buf;
 540         int r = 0;
 541
 542         if (*ppos)
 543                 return 0;
 544
 545         s = kmalloc(sizeof(*s), GFP_KERNEL);
 546         if (!s)
 547                 return -ENOMEM;
 548
 549         trace_seq_init(s);
 550
 551         trace_seq_printf(s, "name: %s\n", call->name);
 552         trace_seq_printf(s, "ID: %d\n", call->id);
 553         trace_seq_printf(s, "format:\n");
 554
 555         list_for_each_entry_reverse(field, &call->fields, link) {
 556                 /*
 557                  * Smartly shows the array type(except dynamic array).
 558                  * Normal:
 559                  *      field:TYPE VAR
 560                  * If TYPE := TYPE[LEN], it is shown:
 561                  *      field:TYPE VAR[LEN]
 562                  */
 563                 const char *array_descriptor = strchr(field->type, '[');
 564
 565                 if (!strncmp(field->type, "__data_loc", 10))
 566                         array_descriptor = NULL;
 567
 568                 if (!array_descriptor) {
 569                         r = trace_seq_printf(s, "\tfield:%s %s;\toffset:%u;"
 570                                         "\tsize:%u;\tsigned:%d;\n",
 571                                         field->type, field->name, field->offset,
 572                                         field->size, !!field->is_signed);
 573                 } else {
 574                         r = trace_seq_printf(s, "\tfield:%.*s %s%s;\toffset:%u;"
 575                                         "\tsize:%u;\tsigned:%d;\n",
 576                                         (int)(array_descriptor - field->type),
 577                                         field->type, field->name,
 578                                         array_descriptor, field->offset,
 579                                         field->size, !!field->is_signed);
 580                 }
 581
 582                 if (--common_field_count == 0)
 583                         r = trace_seq_printf(s, "\n");
 584
 585                 if (!r)
 586                         break;
 587         }
 588
 589         if (r)
 590                 r = trace_seq_printf(s, "\nprint fmt: %s\n",
 591                                 call->print_fmt);
 592
 593         if (!r) {
 594                 /*
 595                  * ug!  The format output is bigger than a PAGE!!
 596                  */
 597                 buf = "FORMAT TOO BIG\n";
 598                 r = simple_read_from_buffer(ubuf, cnt, ppos,
 599                                               buf, strlen(buf));
 600                 goto out;
 601         }
 602
 603         r = simple_read_from_buffer(ubuf, cnt, ppos,
 604                                     s->buffer, s->len);
 605  out:
 606         kfree(s);
 607         return r;
 608 }
 609
 610 static ssize_t
 611 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 612 {
 613         struct ftrace_event_call *call = filp->private_data;
 614         struct trace_seq *s;
 615         int r;
 616
 617         if (*ppos)
 618                 return 0;
 619
 620         s = kmalloc(sizeof(*s), GFP_KERNEL);
 621         if (!s)
 622                 return -ENOMEM;
 623
 624         trace_seq_init(s);
 625         trace_seq_printf(s, "%d\n", call->id);
 626
 627         r = simple_read_from_buffer(ubuf, cnt, ppos,
 628                                     s->buffer, s->len);
 629         kfree(s);
 630         return r;
 631 }
 632
 633 static ssize_t
 634 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 635                   loff_t *ppos)
 636 {
 637         struct ftrace_event_call *call = filp->private_data;
 638         struct trace_seq *s;
 639         int r;
 640
 641         if (*ppos)
 642                 return 0;
 643
 644         s = kmalloc(sizeof(*s), GFP_KERNEL);
 645         if (!s)
 646                 return -ENOMEM;
 647
 648         trace_seq_init(s);
 649
 650         print_event_filter(call, s);
 651         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 652
 653         kfree(s);
 654
 655         return r;
 656 }
 657
 658 static ssize_t
 659 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
 660                    loff_t *ppos)
 661 {
 662         struct ftrace_event_call *call = filp->private_data;
 663         char *buf;
 664         int err;
 665
 666         if (cnt >= PAGE_SIZE)
 667                 return -EINVAL;
 668
 669         buf = (char *)__get_free_page(GFP_TEMPORARY);
 670         if (!buf)
 671                 return -ENOMEM;
 672
 673         if (copy_from_user(buf, ubuf, cnt)) {
 674                 free_page((unsigned long) buf);
 675                 return -EFAULT;
 676         }
 677         buf[cnt] = '\0';
 678
 679         err = apply_event_filter(call, buf);
 680         free_page((unsigned long) buf);
 681         if (err < 0)
 682                 return err;
 683
 684         *ppos += cnt;
 685
 686         return cnt;
 687 }
 688
 689 static ssize_t
 690 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 691                       loff_t *ppos)
 692 {
 693         struct event_subsystem *system = filp->private_data;
 694         struct trace_seq *s;
 695         int r;
 696
 697         if (*ppos)
 698                 return 0;
 699
 700         s = kmalloc(sizeof(*s), GFP_KERNEL);
 701         if (!s)
 702                 return -ENOMEM;
 703
 704         trace_seq_init(s);
 705
 706         print_subsystem_event_filter(system, s);
 707         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 708
 709         kfree(s);
 710
 711         return r;
 712 }
 713
 714 static ssize_t
 715 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
 716                        loff_t *ppos)
 717 {
 718         struct event_subsystem *system = filp->private_data;
 719         char *buf;
 720         int err;
 721
 722         if (cnt >= PAGE_SIZE)
 723                 return -EINVAL;
 724
 725         buf = (char *)__get_free_page(GFP_TEMPORARY);
 726         if (!buf)
 727                 return -ENOMEM;
 728
 729         if (copy_from_user(buf, ubuf, cnt)) {
 730                 free_page((unsigned long) buf);
 731                 return -EFAULT;
 732         }
 733         buf[cnt] = '\0';
 734
 735         err = apply_subsystem_event_filter(system, buf);
 736         free_page((unsigned long) buf);
 737         if (err < 0)
 738                 return err;
 739
 740         *ppos += cnt;
 741
 742         return cnt;
 743 }
 744
 745 static ssize_t
 746 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 747 {
 748         int (*func)(struct trace_seq *s) = filp->private_data;
 749         struct trace_seq *s;
 750         int r;
 751
 752         if (*ppos)
 753                 return 0;
 754
 755         s = kmalloc(sizeof(*s), GFP_KERNEL);
 756         if (!s)
 757                 return -ENOMEM;
 758
 759         trace_seq_init(s);
 760
 761         func(s);
 762         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 763
 764         kfree(s);
 765
 766         return r;
 767 }
 768
 769 static const struct seq_operations show_event_seq_ops = {
 770         .start = t_start,
 771         .next = t_next,
 772         .show = t_show,
 773         .stop = t_stop,
 774 };
 775
 776 static const struct seq_operations show_set_event_seq_ops = {
 777         .start = s_start,
 778         .next = s_next,
 779         .show = t_show,
 780         .stop = t_stop,
 781 };
 782
 783 static const struct file_operations ftrace_avail_fops = {
 784         .open = ftrace_event_seq_open,
 785         .read = seq_read,
 786         .llseek = seq_lseek,
 787         .release = seq_release,
 788 };
 789
 790 static const struct file_operations ftrace_set_event_fops = {
 791         .open = ftrace_event_seq_open,
 792         .read = seq_read,
 793         .write = ftrace_event_write,
 794         .llseek = seq_lseek,
 795         .release = seq_release,
 796 };
 797
 798 static const struct file_operations ftrace_enable_fops = {
 799         .open = tracing_open_generic,
 800         .read = event_enable_read,
 801         .write = event_enable_write,
 802 };
 803
 804 static const struct file_operations ftrace_event_format_fops = {
 805         .open = tracing_open_generic,
 806         .read = event_format_read,
 807 };
 808
 809 static const struct file_operations ftrace_event_id_fops = {
 810         .open = tracing_open_generic,
 811         .read = event_id_read,
 812 };
 813
 814 static const struct file_operations ftrace_event_filter_fops = {
 815         .open = tracing_open_generic,
 816         .read = event_filter_read,
 817         .write = event_filter_write,
 818 };
 819
 820 static const struct file_operations ftrace_subsystem_filter_fops = {
 821         .open = tracing_open_generic,
 822         .read = subsystem_filter_read,
 823         .write = subsystem_filter_write,
 824 };
 825
 826 static const struct file_operations ftrace_system_enable_fops = {
 827         .open = tracing_open_generic,
 828         .read = system_enable_read,
 829         .write = system_enable_write,
 830 };
 831
 832 static const struct file_operations ftrace_show_header_fops = {
 833         .open = tracing_open_generic,
 834         .read = show_header,
 835 };
 836
 837 static struct dentry *event_trace_events_dir(void)
 838 {
 839         static struct dentry *d_tracer;
 840         static struct dentry *d_events;
 841
 842         if (d_events)
 843                 return d_events;
 844
 845         d_tracer = tracing_init_dentry();
 846         if (!d_tracer)
 847                 return NULL;
 848
 849         d_events = debugfs_create_dir("events", d_tracer);
 850         if (!d_events)
 851                 pr_warning("Could not create debugfs "
 852                            "'events' directory\n");
 853
 854         return d_events;
 855 }
 856
 857 static LIST_HEAD(event_subsystems);
 858
 859 static struct dentry *
 860 event_subsystem_dir(const char *name, struct dentry *d_events)
 861 {
 862         struct event_subsystem *system;
 863         struct dentry *entry;
 864
 865         /* First see if we did not already create this dir */
 866         list_for_each_entry(system, &event_subsystems, list) {
 867                 if (strcmp(system->name, name) == 0) {
 868                         system->nr_events++;
 869                         return system->entry;
 870                 }
 871         }
 872
 873         /* need to create new entry */
 874         system = kmalloc(sizeof(*system), GFP_KERNEL);
 875         if (!system) {
 876                 pr_warning("No memory to create event subsystem %s\n",
 877                            name);
 878                 return d_events;
 879         }
 880
 881         system->entry = debugfs_create_dir(name, d_events);
 882         if (!system->entry) {
 883                 pr_warning("Could not create event subsystem %s\n",
 884                            name);
 885                 kfree(system);
 886                 return d_events;
 887         }
 888
 889         system->nr_events = 1;
 890         system->name = kstrdup(name, GFP_KERNEL);
 891         if (!system->name) {
 892                 debugfs_remove(system->entry);
 893                 kfree(system);
 894                 return d_events;
 895         }
 896
 897         list_add(&system->list, &event_subsystems);
 898
 899         system->filter = NULL;
 900
 901         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
 902         if (!system->filter) {
 903                 pr_warning("Could not allocate filter for subsystem "
 904                            "'%s'\n", name);
 905                 return system->entry;
 906         }
 907
 908         entry = debugfs_create_file("filter", 0644, system->entry, system,
 909                                     &ftrace_subsystem_filter_fops);
 910         if (!entry) {
 911                 kfree(system->filter);
 912                 system->filter = NULL;
 913                 pr_warning("Could not create debugfs "
 914                            "'%s/filter' entry\n", name);
 915         }
 916
 917         trace_create_file("enable", 0644, system->entry,
 918                           (void *)system->name,
 919                           &ftrace_system_enable_fops);
 920
 921         return system->entry;
 922 }
 923
 924 static int
 925 event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
 926                  const struct file_operations *id,
 927                  const struct file_operations *enable,
 928                  const struct file_operations *filter,
 929                  const struct file_operations *format)
 930 {
 931         int ret;
 932
 933         /*
 934          * If the trace point header did not define TRACE_SYSTEM
 935          * then the system would be called "TRACE_SYSTEM".
 936          */
 937         if (strcmp(call->system, TRACE_SYSTEM) != 0)
 938                 d_events = event_subsystem_dir(call->system, d_events);
 939
 940         call->dir = debugfs_create_dir(call->name, d_events);
 941         if (!call->dir) {
 942                 pr_warning("Could not create debugfs "
 943                            "'%s' directory\n", call->name);
 944                 return -1;
 945         }
 946
 947         if (call->regfunc)
 948                 trace_create_file("enable", 0644, call->dir, call,
 949                                   enable);
 950
 951         if (call->id && call->profile_enable)
 952                 trace_create_file("id", 0444, call->dir, call,
 953                                   id);
 954
 955         if (call->define_fields) {
 956                 ret = trace_define_common_fields(call);
 957                 if (!ret)
 958                         ret = call->define_fields(call);
 959                 if (ret < 0) {
 960                         pr_warning("Could not initialize trace point"
 961                                    " events/%s\n", call->name);
 962                         return ret;
 963                 }
 964                 trace_create_file("filter", 0644, call->dir, call,
 965                                   filter);
 966         }
 967
 968         /* A trace may not want to export its format */
 969         if (!call->show_format)
 970                 return 0;
 971
 972         trace_create_file("format", 0444, call->dir, call,
 973                           format);
 974
 975         return 0;
 976 }
 977
 978 static int __trace_add_event_call(struct ftrace_event_call *call)
 979 {
 980         struct dentry *d_events;
 981         int ret;
 982
 983         if (!call->name)
 984                 return -EINVAL;
 985
 986         if (call->raw_init) {
 987                 ret = call->raw_init(call);
 988                 if (ret < 0) {
 989                         if (ret != -ENOSYS)
 990                                 pr_warning("Could not initialize trace "
 991                                 "events/%s\n", call->name);
 992                         return ret;
 993                 }
 994         }
 995
 996         d_events = event_trace_events_dir();
 997         if (!d_events)
 998                 return -ENOENT;
 999
1000         ret = event_create_dir(call, d_events, &ftrace_event_id_fops,
1001                                 &ftrace_enable_fops, &ftrace_event_filter_fops,
1002                                 &ftrace_event_format_fops);
1003         if (!ret)
1004                 list_add(&call->list, &ftrace_events);
1005
1006         return ret;
1007 }
1008
1009 /* Add an additional event_call dynamically */
1010 int trace_add_event_call(struct ftrace_event_call *call)
1011 {
1012         int ret;
1013         mutex_lock(&event_mutex);
1014         ret = __trace_add_event_call(call);
1015         mutex_unlock(&event_mutex);
1016         return ret;
1017 }
1018
1019 static void remove_subsystem_dir(const char *name)
1020 {
1021         struct event_subsystem *system;
1022
1023         if (strcmp(name, TRACE_SYSTEM) == 0)
1024                 return;
1025
1026         list_for_each_entry(system, &event_subsystems, list) {
1027                 if (strcmp(system->name, name) == 0) {
1028                         if (!--system->nr_events) {
1029                                 struct event_filter *filter = system->filter;
1030
1031                                 debugfs_remove_recursive(system->entry);
1032                                 list_del(&system->list);
1033                                 if (filter) {
1034                                         kfree(filter->filter_string);
1035                                         kfree(filter);
1036                                 }
1037                                 kfree(system->name);
1038                                 kfree(system);
1039                         }
1040                         break;
1041                 }
1042         }
1043 }
1044
1045 /*
1046  * Must be called under locking both of event_mutex and trace_event_mutex.
1047  */
1048 static void __trace_remove_event_call(struct ftrace_event_call *call)
1049 {
1050         ftrace_event_enable_disable(call, 0);
1051         if (call->event)
1052                 __unregister_ftrace_event(call->event);
1053         debugfs_remove_recursive(call->dir);
1054         list_del(&call->list);
1055         trace_destroy_fields(call);
1056         destroy_preds(call);
1057         remove_subsystem_dir(call->system);
1058 }
1059
1060 /* Remove an event_call */
1061 void trace_remove_event_call(struct ftrace_event_call *call)
1062 {
1063         mutex_lock(&event_mutex);
1064         down_write(&trace_event_mutex);
1065         __trace_remove_event_call(call);
1066         up_write(&trace_event_mutex);
1067         mutex_unlock(&event_mutex);
1068 }
1069
1070 #define for_each_event(event, start, end)                       \
1071         for (event = start;                                     \
1072              (unsigned long)event < (unsigned long)end;         \
1073              event++)
1074
1075 #ifdef CONFIG_MODULES
1076
1077 static LIST_HEAD(ftrace_module_file_list);
1078
1079 /*
1080  * Modules must own their file_operations to keep up with
1081  * reference counting.
1082  */
1083 struct ftrace_module_file_ops {
1084         struct list_head                list;
1085         struct module                   *mod;
1086         struct file_operations          id;
1087         struct file_operations          enable;
1088         struct file_operations          format;
1089         struct file_operations          filter;
1090 };
1091
1092 static struct ftrace_module_file_ops *
1093 trace_create_file_ops(struct module *mod)
1094 {
1095         struct ftrace_module_file_ops *file_ops;
1096
1097         /*
1098          * This is a bit of a PITA. To allow for correct reference
1099          * counting, modules must "own" their file_operations.
1100          * To do this, we allocate the file operations that will be
1101          * used in the event directory.
1102          */
1103
1104         file_ops = kmalloc(sizeof(*file_ops), GFP_KERNEL);
1105         if (!file_ops)
1106                 return NULL;
1107
1108         file_ops->mod = mod;
1109
1110         file_ops->id = ftrace_event_id_fops;
1111         file_ops->id.owner = mod;
1112
1113         file_ops->enable = ftrace_enable_fops;
1114         file_ops->enable.owner = mod;
1115
1116         file_ops->filter = ftrace_event_filter_fops;
1117         file_ops->filter.owner = mod;
1118
1119         file_ops->format = ftrace_event_format_fops;
1120         file_ops->format.owner = mod;
1121
1122         list_add(&file_ops->list, &ftrace_module_file_list);
1123
1124         return file_ops;
1125 }
1126
1127 static void trace_module_add_events(struct module *mod)
1128 {
1129         struct ftrace_module_file_ops *file_ops = NULL;
1130         struct ftrace_event_call *call, *start, *end;
1131         struct dentry *d_events;
1132         int ret;
1133
1134         start = mod->trace_events;
1135         end = mod->trace_events + mod->num_trace_events;
1136
1137         if (start == end)
1138                 return;
1139
1140         d_events = event_trace_events_dir();
1141         if (!d_events)
1142                 return;
1143
1144         for_each_event(call, start, end) {
1145                 /* The linker may leave blanks */
1146                 if (!call->name)
1147                         continue;
1148                 if (call->raw_init) {
1149                         ret = call->raw_init(call);
1150                         if (ret < 0) {
1151                                 if (ret != -ENOSYS)
1152                                         pr_warning("Could not initialize trace "
1153                                         "point events/%s\n", call->name);
1154                                 continue;
1155                         }
1156                 }
1157                 /*
1158                  * This module has events, create file ops for this module
1159                  * if not already done.
1160                  */
1161                 if (!file_ops) {
1162                         file_ops = trace_create_file_ops(mod);
1163                         if (!file_ops)
1164                                 return;
1165                 }
1166                 call->mod = mod;
1167                 ret = event_create_dir(call, d_events,
1168                                        &file_ops->id, &file_ops->enable,
1169                                        &file_ops->filter, &file_ops->format);
1170                 if (!ret)
1171                         list_add(&call->list, &ftrace_events);
1172         }
1173 }
1174
1175 static void trace_module_remove_events(struct module *mod)
1176 {
1177         struct ftrace_module_file_ops *file_ops;
1178         struct ftrace_event_call *call, *p;
1179         bool found = false;
1180
1181         down_write(&trace_event_mutex);
1182         list_for_each_entry_safe(call, p, &ftrace_events, list) {
1183                 if (call->mod == mod) {
1184                         found = true;
1185                         __trace_remove_event_call(call);
1186                 }
1187         }
1188
1189         /* Now free the file_operations */
1190         list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1191                 if (file_ops->mod == mod)
1192                         break;
1193         }
1194         if (&file_ops->list != &ftrace_module_file_list) {
1195                 list_del(&file_ops->list);
1196                 kfree(file_ops);
1197         }
1198
1199         /*
1200          * It is safest to reset the ring buffer if the module being unloaded
1201          * registered any events.
1202          */
1203         if (found)
1204                 tracing_reset_current_online_cpus();
1205         up_write(&trace_event_mutex);
1206 }
1207
1208 static int trace_module_notify(struct notifier_block *self,
1209                                unsigned long val, void *data)
1210 {
1211         struct module *mod = data;
1212
1213         mutex_lock(&event_mutex);
1214         switch (val) {
1215         case MODULE_STATE_COMING:
1216                 trace_module_add_events(mod);
1217                 break;
1218         case MODULE_STATE_GOING:
1219                 trace_module_remove_events(mod);
1220                 break;
1221         }
1222         mutex_unlock(&event_mutex);
1223
1224         return 0;
1225 }
1226 #else
1227 static int trace_module_notify(struct notifier_block *self,
1228                                unsigned long val, void *data)
1229 {
1230         return 0;
1231 }
1232 #endif /* CONFIG_MODULES */
1233
1234 static struct notifier_block trace_module_nb = {
1235         .notifier_call = trace_module_notify,
1236         .priority = 0,
1237 };
1238
1239 extern struct ftrace_event_call __start_ftrace_events[];
1240 extern struct ftrace_event_call __stop_ftrace_events[];
1241
1242 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
1243
1244 static __init int setup_trace_event(char *str)
1245 {
1246         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
1247         ring_buffer_expanded = 1;
1248         tracing_selftest_disabled = 1;
1249
1250         return 1;
1251 }
1252 __setup("trace_event=", setup_trace_event);
1253
1254 static __init int event_trace_init(void)
1255 {
1256         struct ftrace_event_call *call;
1257         struct dentry *d_tracer;
1258         struct dentry *entry;
1259         struct dentry *d_events;
1260         int ret;
1261         char *buf = bootup_event_buf;
1262         char *token;
1263
1264         d_tracer = tracing_init_dentry();
1265         if (!d_tracer)
1266                 return 0;
1267
1268         entry = debugfs_create_file("available_events", 0444, d_tracer,
1269                                     (void *)&show_event_seq_ops,
1270                                     &ftrace_avail_fops);
1271         if (!entry)
1272                 pr_warning("Could not create debugfs "
1273                            "'available_events' entry\n");
1274
1275         entry = debugfs_create_file("set_event", 0644, d_tracer,
1276                                     (void *)&show_set_event_seq_ops,
1277                                     &ftrace_set_event_fops);
1278         if (!entry)
1279                 pr_warning("Could not create debugfs "
1280                            "'set_event' entry\n");
1281
1282         d_events = event_trace_events_dir();
1283         if (!d_events)
1284                 return 0;
1285
1286         /* ring buffer internal formats */
1287         trace_create_file("header_page", 0444, d_events,
1288                           ring_buffer_print_page_header,
1289                           &ftrace_show_header_fops);
1290
1291         trace_create_file("header_event", 0444, d_events,
1292                           ring_buffer_print_entry_header,
1293                           &ftrace_show_header_fops);
1294
1295         trace_create_file("enable", 0644, d_events,
1296                           NULL, &ftrace_system_enable_fops);
1297
1298         for_each_event(call, __start_ftrace_events, __stop_ftrace_events) {
1299                 /* The linker may leave blanks */
1300                 if (!call->name)
1301                         continue;
1302                 if (call->raw_init) {
1303                         ret = call->raw_init(call);
1304                         if (ret < 0) {
1305                                 if (ret != -ENOSYS)
1306                                         pr_warning("Could not initialize trace "
1307                                         "point events/%s\n", call->name);
1308                                 continue;
1309                         }
1310                 }
1311                 ret = event_create_dir(call, d_events, &ftrace_event_id_fops,
1312                                        &ftrace_enable_fops,
1313                                        &ftrace_event_filter_fops,
1314                                        &ftrace_event_format_fops);
1315                 if (!ret)
1316                         list_add(&call->list, &ftrace_events);
1317         }
1318
1319         while (true) {
1320                 token = strsep(&buf, ",");
1321
1322                 if (!token)
1323                         break;
1324                 if (!*token)
1325                         continue;
1326
1327                 ret = ftrace_set_clr_event(token, 1);
1328                 if (ret)
1329                         pr_warning("Failed to enable trace event: %s\n", token);
1330         }
1331
1332         ret = register_module_notifier(&trace_module_nb);
1333         if (ret)
1334                 pr_warning("Failed to register trace events module notifier\n");
1335
1336         return 0;
1337 }
1338 fs_initcall(event_trace_init);
1339
1340 #ifdef CONFIG_FTRACE_STARTUP_TEST
1341
1342 static DEFINE_SPINLOCK(test_spinlock);
1343 static DEFINE_SPINLOCK(test_spinlock_irq);
1344 static DEFINE_MUTEX(test_mutex);
1345
1346 static __init void test_work(struct work_struct *dummy)
1347 {
1348         spin_lock(&test_spinlock);
1349         spin_lock_irq(&test_spinlock_irq);
1350         udelay(1);
1351         spin_unlock_irq(&test_spinlock_irq);
1352         spin_unlock(&test_spinlock);
1353
1354         mutex_lock(&test_mutex);
1355         msleep(1);
1356         mutex_unlock(&test_mutex);
1357 }
1358
1359 static __init int event_test_thread(void *unused)
1360 {
1361         void *test_malloc;
1362
1363         test_malloc = kmalloc(1234, GFP_KERNEL);
1364         if (!test_malloc)
1365                 pr_info("failed to kmalloc\n");
1366
1367         schedule_on_each_cpu(test_work);
1368
1369         kfree(test_malloc);
1370
1371         set_current_state(TASK_INTERRUPTIBLE);
1372         while (!kthread_should_stop())
1373                 schedule();
1374
1375         return 0;
1376 }
1377
1378 /*
1379  * Do various things that may trigger events.
1380  */
1381 static __init void event_test_stuff(void)
1382 {
1383         struct task_struct *test_thread;
1384
1385         test_thread = kthread_run(event_test_thread, NULL, "test-events");
1386         msleep(1);
1387         kthread_stop(test_thread);
1388 }
1389
1390 /*
1391  * For every trace event defined, we will test each trace point separately,
1392  * and then by groups, and finally all trace points.
1393  */
1394 static __init void event_trace_self_tests(void)
1395 {
1396         struct ftrace_event_call *call;
1397         struct event_subsystem *system;
1398         int ret;
1399
1400         pr_info("Running tests on trace events:\n");
1401
1402         list_for_each_entry(call, &ftrace_events, list) {
1403
1404                 /* Only test those that have a regfunc */
1405                 if (!call->regfunc)
1406                         continue;
1407
1408 /*
1409  * Testing syscall events here is pretty useless, but
1410  * we still do it if configured. But this is time consuming.
1411  * What we really need is a user thread to perform the
1412  * syscalls as we test.
1413  */
1414 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
1415                 if (call->system &&
1416                     strcmp(call->system, "syscalls") == 0)
1417                         continue;
1418 #endif
1419
1420                 pr_info("Testing event %s: ", call->name);
1421
1422                 /*
1423                  * If an event is already enabled, someone is using
1424                  * it and the self test should not be on.
1425                  */
1426                 if (call->enabled) {
1427                         pr_warning("Enabled event during self test!\n");
1428                         WARN_ON_ONCE(1);
1429                         continue;
1430                 }
1431
1432                 ftrace_event_enable_disable(call, 1);
1433                 event_test_stuff();
1434                 ftrace_event_enable_disable(call, 0);
1435
1436                 pr_cont("OK\n");
1437         }
1438
1439         /* Now test at the sub system level */
1440
1441         pr_info("Running tests on trace event systems:\n");
1442
1443         list_for_each_entry(system, &event_subsystems, list) {
1444
1445                 /* the ftrace system is special, skip it */
1446                 if (strcmp(system->name, "ftrace") == 0)
1447                         continue;
1448
1449                 pr_info("Testing event system %s: ", system->name);
1450
1451                 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 1);
1452                 if (WARN_ON_ONCE(ret)) {
1453                         pr_warning("error enabling system %s\n",
1454                                    system->name);
1455                         continue;
1456                 }
1457
1458                 event_test_stuff();
1459
1460                 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 0);
1461                 if (WARN_ON_ONCE(ret))
1462                         pr_warning("error disabling system %s\n",
1463                                    system->name);
1464
1465                 pr_cont("OK\n");
1466         }
1467
1468         /* Test with all events enabled */
1469
1470         pr_info("Running tests on all trace events:\n");
1471         pr_info("Testing all events: ");
1472
1473         ret = __ftrace_set_clr_event(NULL, NULL, NULL, 1);
1474         if (WARN_ON_ONCE(ret)) {
1475                 pr_warning("error enabling all events\n");
1476                 return;
1477         }
1478
1479         event_test_stuff();
1480
1481         /* reset sysname */
1482         ret = __ftrace_set_clr_event(NULL, NULL, NULL, 0);
1483         if (WARN_ON_ONCE(ret)) {
1484                 pr_warning("error disabling all events\n");
1485                 return;
1486         }
1487
1488         pr_cont("OK\n");
1489 }
1490
1491 #ifdef CONFIG_FUNCTION_TRACER
1492
1493 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
1494
1495 static void
1496 function_test_events_call(unsigned long ip, unsigned long parent_ip)
1497 {
1498         struct ring_buffer_event *event;
1499         struct ring_buffer *buffer;
1500         struct ftrace_entry *entry;
1501         unsigned long flags;
1502         long disabled;
1503         int resched;
1504         int cpu;
1505         int pc;
1506
1507         pc = preempt_count();
1508         resched = ftrace_preempt_disable();
1509         cpu = raw_smp_processor_id();
1510         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
1511
1512         if (disabled != 1)
1513                 goto out;
1514
1515         local_save_flags(flags);
1516
1517         event = trace_current_buffer_lock_reserve(&buffer,
1518                                                   TRACE_FN, sizeof(*entry),
1519                                                   flags, pc);
1520         if (!event)
1521                 goto out;
1522         entry   = ring_buffer_event_data(event);
1523         entry->ip                       = ip;
1524         entry->parent_ip                = parent_ip;
1525
1526         trace_nowake_buffer_unlock_commit(buffer, event, flags, pc);
1527
1528  out:
1529         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
1530         ftrace_preempt_enable(resched);
1531 }
1532
1533 static struct ftrace_ops trace_ops __initdata  =
1534 {
1535         .func = function_test_events_call,
1536 };
1537
1538 static __init void event_trace_self_test_with_function(void)
1539 {
1540         register_ftrace_function(&trace_ops);
1541         pr_info("Running tests again, along with the function tracer\n");
1542         event_trace_self_tests();
1543         unregister_ftrace_function(&trace_ops);
1544 }
1545 #else
1546 static __init void event_trace_self_test_with_function(void)
1547 {
1548 }
1549 #endif
1550
1551 static __init int event_trace_self_tests_init(void)
1552 {
1553         if (!tracing_selftest_disabled) {
1554                 event_trace_self_tests();
1555                 event_trace_self_test_with_function();
1556         }
1557
1558         return 0;
1559 }
1560
1561 late_initcall(event_trace_self_tests_init);
1562
1563 #endif