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