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