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

   1 /*
   2  * event tracer
   3  *
   4  * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
   5  *
   6  *  - Added format output of fields of the trace point.
   7  *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
   8  *
   9  */
  10
  11 #include <linux/workqueue.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/kthread.h>
  14 #include <linux/debugfs.h>
  15 #include <linux/uaccess.h>
  16 #include <linux/module.h>
  17 #include <linux/ctype.h>
  18 #include <linux/delay.h>
  19
  20 #include <asm/setup.h>
  21
  22 #include "trace_output.h"
  23
  24 #undef TRACE_SYSTEM
  25 #define TRACE_SYSTEM "TRACE_SYSTEM"
  26
  27 DEFINE_MUTEX(event_mutex);
  28
  29 LIST_HEAD(ftrace_events);
  30
  31 int trace_define_field(struct ftrace_event_call *call, const char *type,
  32                        const char *name, int offset, int size, int is_signed,
  33                        int filter_type)
  34 {
  35         struct ftrace_event_field *field;
  36
  37         field = kzalloc(sizeof(*field), GFP_KERNEL);
  38         if (!field)
  39                 goto err;
  40
  41         field->name = kstrdup(name, GFP_KERNEL);
  42         if (!field->name)
  43                 goto err;
  44
  45         field->type = kstrdup(type, GFP_KERNEL);
  46         if (!field->type)
  47                 goto err;
  48
  49         if (filter_type == FILTER_OTHER)
  50                 field->filter_type = filter_assign_type(type);
  51         else
  52                 field->filter_type = filter_type;
  53
  54         field->offset = offset;
  55         field->size = size;
  56         field->is_signed = is_signed;
  57
  58         list_add(&field->link, &call->fields);
  59
  60         return 0;
  61
  62 err:
  63         if (field)
  64                 kfree(field->name);
  65         kfree(field);
  66
  67         return -ENOMEM;
  68 }
  69 EXPORT_SYMBOL_GPL(trace_define_field);
  70
  71 #define __common_field(type, item)                                      \
  72         ret = trace_define_field(call, #type, "common_" #item,          \
  73                                  offsetof(typeof(ent), item),           \
  74                                  sizeof(ent.item),                      \
  75                                  is_signed_type(type), FILTER_OTHER);   \
  76         if (ret)                                                        \
  77                 return ret;
  78
  79 static int trace_define_common_fields(struct ftrace_event_call *call)
  80 {
  81         int ret;
  82         struct trace_entry ent;
  83
  84         __common_field(unsigned short, type);
  85         __common_field(unsigned char, flags);
  86         __common_field(unsigned char, preempt_count);
  87         __common_field(int, pid);
  88         __common_field(int, lock_depth);
  89
  90         return ret;
  91 }
  92
  93 void trace_destroy_fields(struct ftrace_event_call *call)
  94 {
  95         struct ftrace_event_field *field, *next;
  96
  97         list_for_each_entry_safe(field, next, &call->fields, link) {
  98                 list_del(&field->link);
  99                 kfree(field->type);
 100                 kfree(field->name);
 101                 kfree(field);
 102         }
 103 }
 104
 105 int trace_event_raw_init(struct ftrace_event_call *call)
 106 {
 107         int id;
 108
 109         id = register_ftrace_event(call->event);
 110         if (!id)
 111                 return -ENODEV;
 112         call->id = id;
 113         INIT_LIST_HEAD(&call->fields);
 114
 115         return 0;
 116 }
 117 EXPORT_SYMBOL_GPL(trace_event_raw_init);
 118
 119 static int ftrace_event_enable_disable(struct ftrace_event_call *call,
 120                                         int enable)
 121 {
 122         int ret = 0;
 123
 124         switch (enable) {
 125         case 0:
 126                 if (call->enabled) {
 127                         call->enabled = 0;
 128                         tracing_stop_cmdline_record();
 129                         call->unregfunc(call);
 130                 }
 131                 break;
 132         case 1:
 133                 if (!call->enabled) {
 134                         tracing_start_cmdline_record();
 135                         ret = call->regfunc(call);
 136                         if (ret) {
 137                                 tracing_stop_cmdline_record();
 138                                 pr_info("event trace: Could not enable event "
 139                                         "%s\n", call->name);
 140                                 break;
 141                         }
 142                         call->enabled = 1;
 143                 }
 144                 break;
 145         }
 146
 147         return ret;
 148 }
 149
 150 static void ftrace_clear_events(void)
 151 {
 152         struct ftrace_event_call *call;
 153
 154         mutex_lock(&event_mutex);
 155         list_for_each_entry(call, &ftrace_events, list) {
 156                 ftrace_event_enable_disable(call, 0);
 157         }
 158         mutex_unlock(&event_mutex);
 159 }
 160
 161 /*
 162  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 163  */
 164 static int __ftrace_set_clr_event(const char *match, const char *sub,
 165                                   const char *event, int set)
 166 {
 167         struct ftrace_event_call *call;
 168         int ret = -EINVAL;
 169
 170         mutex_lock(&event_mutex);
 171         list_for_each_entry(call, &ftrace_events, list) {
 172
 173                 if (!call->name || !call->regfunc)
 174                         continue;
 175
 176                 if (match &&
 177                     strcmp(match, call->name) != 0 &&
 178                     strcmp(match, call->system) != 0)
 179                         continue;
 180
 181                 if (sub && strcmp(sub, call->system) != 0)
 182                         continue;
 183
 184                 if (event && strcmp(event, call->name) != 0)
 185                         continue;
 186
 187                 ftrace_event_enable_disable(call, set);
 188
 189                 ret = 0;
 190         }
 191         mutex_unlock(&event_mutex);
 192
 193         return ret;
 194 }
 195
 196 static int ftrace_set_clr_event(char *buf, int set)
 197 {
 198         char *event = NULL, *sub = NULL, *match;
 199
 200         /*
 201          * The buf format can be <subsystem>:<event-name>
 202          *  *:<event-name> means any event by that name.
 203          *  :<event-name> is the same.
 204          *
 205          *  <subsystem>:* means all events in that subsystem
 206          *  <subsystem>: means the same.
 207          *
 208          *  <name> (no ':') means all events in a subsystem with
 209          *  the name <name> or any event that matches <name>
 210          */
 211
 212         match = strsep(&buf, ":");
 213         if (buf) {
 214                 sub = match;
 215                 event = buf;
 216                 match = NULL;
 217
 218                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 219                         sub = NULL;
 220                 if (!strlen(event) || strcmp(event, "*") == 0)
 221                         event = NULL;
 222         }
 223
 224         return __ftrace_set_clr_event(match, sub, event, set);
 225 }
 226
 227 /**
 228  * trace_set_clr_event - enable or disable an event
 229  * @system: system name to match (NULL for any system)
 230  * @event: event name to match (NULL for all events, within system)
 231  * @set: 1 to enable, 0 to disable
 232  *
 233  * This is a way for other parts of the kernel to enable or disable
 234  * event recording.
 235  *
 236  * Returns 0 on success, -EINVAL if the parameters do not match any
 237  * registered events.
 238  */
 239 int trace_set_clr_event(const char *system, const char *event, int set)
 240 {
 241         return __ftrace_set_clr_event(NULL, system, event, set);
 242 }
 243
 244 /* 128 should be much more than enough */
 245 #define EVENT_BUF_SIZE          127
 246
 247 static ssize_t
 248 ftrace_event_write(struct file *file, const char __user *ubuf,
 249                    size_t cnt, loff_t *ppos)
 250 {
 251         struct trace_parser parser;
 252         ssize_t read, ret;
 253
 254         if (!cnt)
 255                 return 0;
 256
 257         ret = tracing_update_buffers();
 258         if (ret < 0)
 259                 return ret;
 260
 261         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 262                 return -ENOMEM;
 263
 264         read = trace_get_user(&parser, ubuf, cnt, ppos);
 265
 266         if (read >= 0 && trace_parser_loaded((&parser))) {
 267                 int set = 1;
 268
 269                 if (*parser.buffer == '!')
 270                         set = 0;
 271
 272                 parser.buffer[parser.idx] = 0;
 273
 274                 ret = ftrace_set_clr_event(parser.buffer + !set, set);
 275                 if (ret)
 276                         goto out_put;
 277         }
 278
 279         ret = read;
 280
 281  out_put:
 282         trace_parser_put(&parser);
 283
 284         return ret;
 285 }
 286
 287 static void *
 288 t_next(struct seq_file *m, void *v, loff_t *pos)
 289 {
 290         struct ftrace_event_call *call = v;
 291
 292         (*pos)++;
 293
 294         list_for_each_entry_continue(call, &ftrace_events, list) {
 295                 /*
 296                  * The ftrace subsystem is for showing formats only.
 297                  * They can not be enabled or disabled via the event files.
 298                  */
 299                 if (call->regfunc)
 300                         return call;
 301         }
 302
 303         return NULL;
 304 }
 305
 306 static void *t_start(struct seq_file *m, loff_t *pos)
 307 {
 308         struct ftrace_event_call *call;
 309         loff_t l;
 310
 311         mutex_lock(&event_mutex);
 312
 313         call = list_entry(&ftrace_events, struct ftrace_event_call, list);
 314         for (l = 0; l <= *pos; ) {
 315                 call = t_next(m, call, &l);
 316                 if (!call)
 317                         break;
 318         }
 319         return call;
 320 }
 321
 322 static void *
 323 s_next(struct seq_file *m, void *v, loff_t *pos)
 324 {
 325         struct ftrace_event_call *call = v;
 326
 327         (*pos)++;
 328
 329         list_for_each_entry_continue(call, &ftrace_events, list) {
 330                 if (call->enabled)
 331                         return call;
 332         }
 333
 334         return NULL;
 335 }
 336
 337 static void *s_start(struct seq_file *m, loff_t *pos)
 338 {
 339         struct ftrace_event_call *call;
 340         loff_t l;
 341
 342         mutex_lock(&event_mutex);
 343
 344         call = list_entry(&ftrace_events, struct ftrace_event_call, list);
 345         for (l = 0; l <= *pos; ) {
 346                 call = s_next(m, call, &l);
 347                 if (!call)
 348                         break;
 349         }
 350         return call;
 351 }
 352
 353 static int t_show(struct seq_file *m, void *v)
 354 {
 355         struct ftrace_event_call *call = v;
 356
 357         if (strcmp(call->system, TRACE_SYSTEM) != 0)
 358                 seq_printf(m, "%s:", call->system);
 359         seq_printf(m, "%s\n", call->name);
 360
 361         return 0;
 362 }
 363
 364 static void t_stop(struct seq_file *m, void *p)
 365 {
 366         mutex_unlock(&event_mutex);
 367 }
 368
 369 static int
 370 ftrace_event_seq_open(struct inode *inode, struct file *file)
 371 {
 372         const struct seq_operations *seq_ops;
 373
 374         if ((file->f_mode & FMODE_WRITE) &&
 375             (file->f_flags & O_TRUNC))
 376                 ftrace_clear_events();
 377
 378         seq_ops = inode->i_private;
 379         return seq_open(file, seq_ops);
 380 }
 381
 382 static ssize_t
 383 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 384                   loff_t *ppos)
 385 {
 386         struct ftrace_event_call *call = filp->private_data;
 387         char *buf;
 388
 389         if (call->enabled)
 390                 buf = "1\n";
 391         else
 392                 buf = "0\n";
 393
 394         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 395 }
 396
 397 static ssize_t
 398 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 399                    loff_t *ppos)
 400 {
 401         struct ftrace_event_call *call = filp->private_data;
 402         char buf[64];
 403         unsigned long val;
 404         int ret;
 405
 406         if (cnt >= sizeof(buf))
 407                 return -EINVAL;
 408
 409         if (copy_from_user(&buf, ubuf, cnt))
 410                 return -EFAULT;
 411
 412         buf[cnt] = 0;
 413
 414         ret = strict_strtoul(buf, 10, &val);
 415         if (ret < 0)
 416                 return ret;
 417
 418         ret = tracing_update_buffers();
 419         if (ret < 0)
 420                 return ret;
 421
 422         switch (val) {
 423         case 0:
 424         case 1:
 425                 mutex_lock(&event_mutex);
 426                 ret = ftrace_event_enable_disable(call, val);
 427                 mutex_unlock(&event_mutex);
 428                 break;
 429
 430         default:
 431                 return -EINVAL;
 432         }
 433
 434         *ppos += cnt;
 435
 436         return ret ? ret : cnt;
 437 }
 438
 439 static ssize_t
 440 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 441                    loff_t *ppos)
 442 {
 443         const char set_to_char[4] = { '?', '0', '1', 'X' };
 444         const char *system = filp->private_data;
 445         struct ftrace_event_call *call;
 446         char buf[2];
 447         int set = 0;
 448         int ret;
 449
 450         mutex_lock(&event_mutex);
 451         list_for_each_entry(call, &ftrace_events, list) {
 452                 if (!call->name || !call->regfunc)
 453                         continue;
 454
 455                 if (system && strcmp(call->system, system) != 0)
 456                         continue;
 457
 458                 /*
 459                  * We need to find out if all the events are set
 460                  * or if all events or cleared, or if we have
 461                  * a mixture.
 462                  */
 463                 set |= (1 << !!call->enabled);
 464
 465                 /*
 466                  * If we have a mixture, no need to look further.
 467                  */
 468                 if (set == 3)
 469                         break;
 470         }
 471         mutex_unlock(&event_mutex);
 472
 473         buf[0] = set_to_char[set];
 474         buf[1] = '\n';
 475
 476         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 477
 478         return ret;
 479 }
 480
 481 static ssize_t
 482 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 483                     loff_t *ppos)
 484 {
 485         const char *system = filp->private_data;
 486         unsigned long val;
 487         char buf[64];
 488         ssize_t ret;
 489
 490         if (cnt >= sizeof(buf))
 491                 return -EINVAL;
 492
 493         if (copy_from_user(&buf, ubuf, cnt))
 494                 return -EFAULT;
 495
 496         buf[cnt] = 0;
 497
 498         ret = strict_strtoul(buf, 10, &val);
 499         if (ret < 0)
 500                 return ret;
 501
 502         ret = tracing_update_buffers();
 503         if (ret < 0)
 504                 return ret;
 505
 506         if (val != 0 && val != 1)
 507                 return -EINVAL;
 508
 509         ret = __ftrace_set_clr_event(NULL, system, NULL, val);
 510         if (ret)
 511                 goto out;
 512
 513         ret = cnt;
 514
 515 out:
 516         *ppos += cnt;
 517
 518         return ret;
 519 }
 520
 521 static ssize_t
 522 event_format_read(struct file *filp, char __user *ubuf, size_t cnt,
 523                   loff_t *ppos)
 524 {
 525         struct ftrace_event_call *call = filp->private_data;
 526         struct ftrace_event_field *field;
 527         struct trace_seq *s;
 528         int common_field_count = 5;
 529         char *buf;
 530         int r = 0;
 531
 532         if (*ppos)
 533                 return 0;
 534
 535         s = kmalloc(sizeof(*s), GFP_KERNEL);
 536         if (!s)
 537                 return -ENOMEM;
 538
 539         trace_seq_init(s);
 540
 541         trace_seq_printf(s, "name: %s\n", call->name);
 542         trace_seq_printf(s, "ID: %d\n", call->id);
 543         trace_seq_printf(s, "format:\n");
 544
 545         list_for_each_entry_reverse(field, &call->fields, link) {
 546                 /*
 547                  * Smartly shows the array type(except dynamic array).
 548                  * Normal:
 549                  *      field:TYPE VAR
 550                  * If TYPE := TYPE[LEN], it is shown:
 551                  *      field:TYPE VAR[LEN]
 552                  */
 553                 const char *array_descriptor = strchr(field->type, '[');
 554
 555                 if (!strncmp(field->type, "__data_loc", 10))
 556                         array_descriptor = NULL;
 557
 558                 if (!array_descriptor) {
 559                         r = trace_seq_printf(s, "\tfield:%s %s;\toffset:%u;"
 560                                         "\tsize:%u;\tsigned:%d;\n",
 561                                         field->type, field->name, field->offset,
 562                                         field->size, !!field->is_signed);
 563                 } else {
 564                         r = trace_seq_printf(s, "\tfield:%.*s %s%s;\toffset:%u;"
 565                                         "\tsize:%u;\tsigned:%d;\n",
 566                                         (int)(array_descriptor - field->type),
 567                                         field->type, field->name,
 568                                         array_descriptor, field->offset,
 569                                         field->size, !!field->is_signed);
 570                 }
 571
 572                 if (--common_field_count == 0)
 573                         r = trace_seq_printf(s, "\n");
 574
 575                 if (!r)
 576                         break;
 577         }
 578
 579         if (r)
 580                 r = trace_seq_printf(s, "\nprint fmt: %s\n",
 581                                 call->print_fmt);
 582
 583         if (!r) {
 584                 /*
 585                  * ug!  The format output is bigger than a PAGE!!
 586                  */
 587                 buf = "FORMAT TOO BIG\n";
 588                 r = simple_read_from_buffer(ubuf, cnt, ppos,
 589                                               buf, strlen(buf));
 590                 goto out;
 591         }
 592
 593         r = simple_read_from_buffer(ubuf, cnt, ppos,
 594                                     s->buffer, s->len);
 595  out:
 596         kfree(s);
 597         return r;
 598 }
 599
 600 static ssize_t
 601 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 602 {
 603         struct ftrace_event_call *call = filp->private_data;
 604         struct trace_seq *s;
 605         int r;
 606
 607         if (*ppos)
 608                 return 0;
 609
 610         s = kmalloc(sizeof(*s), GFP_KERNEL);
 611         if (!s)
 612                 return -ENOMEM;
 613
 614         trace_seq_init(s);
 615         trace_seq_printf(s, "%d\n", call->id);
 616
 617         r = simple_read_from_buffer(ubuf, cnt, ppos,
 618                                     s->buffer, s->len);
 619         kfree(s);
 620         return r;
 621 }
 622
 623 static ssize_t
 624 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 625                   loff_t *ppos)
 626 {
 627         struct ftrace_event_call *call = filp->private_data;
 628         struct trace_seq *s;
 629         int r;
 630
 631         if (*ppos)
 632                 return 0;
 633
 634         s = kmalloc(sizeof(*s), GFP_KERNEL);
 635         if (!s)
 636                 return -ENOMEM;
 637
 638         trace_seq_init(s);
 639
 640         print_event_filter(call, s);
 641         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 642
 643         kfree(s);
 644
 645         return r;
 646 }
 647
 648 static ssize_t
 649 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
 650                    loff_t *ppos)
 651 {
 652         struct ftrace_event_call *call = filp->private_data;
 653         char *buf;
 654         int err;
 655
 656         if (cnt >= PAGE_SIZE)
 657                 return -EINVAL;
 658
 659         buf = (char *)__get_free_page(GFP_TEMPORARY);
 660         if (!buf)
 661                 return -ENOMEM;
 662
 663         if (copy_from_user(buf, ubuf, cnt)) {
 664                 free_page((unsigned long) buf);
 665                 return -EFAULT;
 666         }
 667         buf[cnt] = '\0';
 668
 669         err = apply_event_filter(call, buf);
 670         free_page((unsigned long) buf);
 671         if (err < 0)
 672                 return err;
 673
 674         *ppos += cnt;
 675
 676         return cnt;
 677 }
 678
 679 static ssize_t
 680 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 681                       loff_t *ppos)
 682 {
 683         struct event_subsystem *system = filp->private_data;
 684         struct trace_seq *s;
 685         int r;
 686
 687         if (*ppos)
 688                 return 0;
 689
 690         s = kmalloc(sizeof(*s), GFP_KERNEL);
 691         if (!s)
 692                 return -ENOMEM;
 693
 694         trace_seq_init(s);
 695
 696         print_subsystem_event_filter(system, s);
 697         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 698
 699         kfree(s);
 700
 701         return r;
 702 }
 703
 704 static ssize_t
 705 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
 706                        loff_t *ppos)
 707 {
 708         struct event_subsystem *system = filp->private_data;
 709         char *buf;
 710         int err;
 711
 712         if (cnt >= PAGE_SIZE)
 713                 return -EINVAL;
 714
 715         buf = (char *)__get_free_page(GFP_TEMPORARY);
 716         if (!buf)
 717                 return -ENOMEM;
 718
 719         if (copy_from_user(buf, ubuf, cnt)) {
 720                 free_page((unsigned long) buf);
 721                 return -EFAULT;
 722         }
 723         buf[cnt] = '\0';
 724
 725         err = apply_subsystem_event_filter(system, buf);
 726         free_page((unsigned long) buf);
 727         if (err < 0)
 728                 return err;
 729
 730         *ppos += cnt;
 731
 732         return cnt;
 733 }
 734
 735 static ssize_t
 736 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 737 {
 738         int (*func)(struct trace_seq *s) = filp->private_data;
 739         struct trace_seq *s;
 740         int r;
 741
 742         if (*ppos)
 743                 return 0;
 744
 745         s = kmalloc(sizeof(*s), GFP_KERNEL);
 746         if (!s)
 747                 return -ENOMEM;
 748
 749         trace_seq_init(s);
 750
 751         func(s);
 752         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 753
 754         kfree(s);
 755
 756         return r;
 757 }
 758
 759 static const struct seq_operations show_event_seq_ops = {
 760         .start = t_start,
 761         .next = t_next,
 762         .show = t_show,
 763         .stop = t_stop,
 764 };
 765
 766 static const struct seq_operations show_set_event_seq_ops = {
 767         .start = s_start,
 768         .next = s_next,
 769         .show = t_show,
 770         .stop = t_stop,
 771 };
 772
 773 static const struct file_operations ftrace_avail_fops = {
 774         .open = ftrace_event_seq_open,
 775         .read = seq_read,
 776         .llseek = seq_lseek,
 777         .release = seq_release,
 778 };
 779
 780 static const struct file_operations ftrace_set_event_fops = {
 781         .open = ftrace_event_seq_open,
 782         .read = seq_read,
 783         .write = ftrace_event_write,
 784         .llseek = seq_lseek,
 785         .release = seq_release,
 786 };
 787
 788 static const struct file_operations ftrace_enable_fops = {
 789         .open = tracing_open_generic,
 790         .read = event_enable_read,
 791         .write = event_enable_write,
 792 };
 793
 794 static const struct file_operations ftrace_event_format_fops = {
 795         .open = tracing_open_generic,
 796         .read = event_format_read,
 797 };
 798
 799 static const struct file_operations ftrace_event_id_fops = {
 800         .open = tracing_open_generic,
 801         .read = event_id_read,
 802 };
 803
 804 static const struct file_operations ftrace_event_filter_fops = {
 805         .open = tracing_open_generic,
 806         .read = event_filter_read,
 807         .write = event_filter_write,
 808 };
 809
 810 static const struct file_operations ftrace_subsystem_filter_fops = {
 811         .open = tracing_open_generic,
 812         .read = subsystem_filter_read,
 813         .write = subsystem_filter_write,
 814 };
 815
 816 static const struct file_operations ftrace_system_enable_fops = {
 817         .open = tracing_open_generic,
 818         .read = system_enable_read,
 819         .write = system_enable_write,
 820 };
 821
 822 static const struct file_operations ftrace_show_header_fops = {
 823         .open = tracing_open_generic,
 824         .read = show_header,
 825 };
 826
 827 static struct dentry *event_trace_events_dir(void)
 828 {
 829         static struct dentry *d_tracer;
 830         static struct dentry *d_events;
 831
 832         if (d_events)
 833                 return d_events;
 834
 835         d_tracer = tracing_init_dentry();
 836         if (!d_tracer)
 837                 return NULL;
 838
 839         d_events = debugfs_create_dir("events", d_tracer);
 840         if (!d_events)
 841                 pr_warning("Could not create debugfs "
 842                            "'events' directory\n");
 843
 844         return d_events;
 845 }
 846
 847 static LIST_HEAD(event_subsystems);
 848
 849 static struct dentry *
 850 event_subsystem_dir(const char *name, struct dentry *d_events)
 851 {
 852         struct event_subsystem *system;
 853         struct dentry *entry;
 854
 855         /* First see if we did not already create this dir */
 856         list_for_each_entry(system, &event_subsystems, list) {
 857                 if (strcmp(system->name, name) == 0) {
 858                         system->nr_events++;
 859                         return system->entry;
 860                 }
 861         }
 862
 863         /* need to create new entry */
 864         system = kmalloc(sizeof(*system), GFP_KERNEL);
 865         if (!system) {
 866                 pr_warning("No memory to create event subsystem %s\n",
 867                            name);
 868                 return d_events;
 869         }
 870
 871         system->entry = debugfs_create_dir(name, d_events);
 872         if (!system->entry) {
 873                 pr_warning("Could not create event subsystem %s\n",
 874                            name);
 875                 kfree(system);
 876                 return d_events;
 877         }
 878
 879         system->nr_events = 1;
 880         system->name = kstrdup(name, GFP_KERNEL);
 881         if (!system->name) {
 882                 debugfs_remove(system->entry);
 883                 kfree(system);
 884                 return d_events;
 885         }
 886
 887         list_add(&system->list, &event_subsystems);
 888
 889         system->filter = NULL;
 890
 891         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
 892         if (!system->filter) {
 893                 pr_warning("Could not allocate filter for subsystem "
 894                            "'%s'\n", name);
 895                 return system->entry;
 896         }
 897
 898         entry = debugfs_create_file("filter", 0644, system->entry, system,
 899                                     &ftrace_subsystem_filter_fops);
 900         if (!entry) {
 901                 kfree(system->filter);
 902                 system->filter = NULL;
 903                 pr_warning("Could not create debugfs "
 904                            "'%s/filter' entry\n", name);
 905         }
 906
 907         trace_create_file("enable", 0644, system->entry,
 908                           (void *)system->name,
 909                           &ftrace_system_enable_fops);
 910
 911         return system->entry;
 912 }
 913
 914 static int
 915 event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
 916                  const struct file_operations *id,
 917                  const struct file_operations *enable,
 918                  const struct file_operations *filter,
 919                  const struct file_operations *format)
 920 {
 921         int ret;
 922
 923         /*
 924          * If the trace point header did not define TRACE_SYSTEM
 925          * then the system would be called "TRACE_SYSTEM".
 926          */
 927         if (strcmp(call->system, TRACE_SYSTEM) != 0)
 928                 d_events = event_subsystem_dir(call->system, d_events);
 929
 930         call->dir = debugfs_create_dir(call->name, d_events);
 931         if (!call->dir) {
 932                 pr_warning("Could not create debugfs "
 933                            "'%s' directory\n", call->name);
 934                 return -1;
 935         }
 936
 937         if (call->regfunc)
 938                 trace_create_file("enable", 0644, call->dir, call,
 939                                   enable);
 940
 941         if (call->id && call->profile_enable)
 942                 trace_create_file("id", 0444, call->dir, call,
 943                                   id);
 944
 945         if (call->define_fields) {
 946                 ret = trace_define_common_fields(call);
 947                 if (!ret)
 948                         ret = call->define_fields(call);
 949                 if (ret < 0) {
 950                         pr_warning("Could not initialize trace point"
 951                                    " events/%s\n", call->name);
 952                         return ret;
 953                 }
 954                 trace_create_file("filter", 0644, call->dir, call,
 955                                   filter);
 956         }
 957
 958         trace_create_file("format", 0444, call->dir, call,
 959                           format);
 960
 961         return 0;
 962 }
 963
 964 static int __trace_add_event_call(struct ftrace_event_call *call)
 965 {
 966         struct dentry *d_events;
 967         int ret;
 968
 969         if (!call->name)
 970                 return -EINVAL;
 971
 972         if (call->raw_init) {
 973                 ret = call->raw_init(call);
 974                 if (ret < 0) {
 975                         if (ret != -ENOSYS)
 976                                 pr_warning("Could not initialize trace "
 977                                 "events/%s\n", call->name);
 978                         return ret;
 979                 }
 980         }
 981
 982         d_events = event_trace_events_dir();
 983         if (!d_events)
 984                 return -ENOENT;
 985
 986         ret = event_create_dir(call, d_events, &ftrace_event_id_fops,
 987                                 &ftrace_enable_fops, &ftrace_event_filter_fops,
 988                                 &ftrace_event_format_fops);
 989         if (!ret)
 990                 list_add(&call->list, &ftrace_events);
 991
 992         return ret;
 993 }
 994
 995 /* Add an additional event_call dynamically */
 996 int trace_add_event_call(struct ftrace_event_call *call)
 997 {
 998         int ret;
 999         mutex_lock(&event_mutex);
1000         ret = __trace_add_event_call(call);
1001         mutex_unlock(&event_mutex);
1002         return ret;
1003 }
1004
1005 static void remove_subsystem_dir(const char *name)
1006 {
1007         struct event_subsystem *system;
1008
1009         if (strcmp(name, TRACE_SYSTEM) == 0)
1010                 return;
1011
1012         list_for_each_entry(system, &event_subsystems, list) {
1013                 if (strcmp(system->name, name) == 0) {
1014                         if (!--system->nr_events) {
1015                                 struct event_filter *filter = system->filter;
1016
1017                                 debugfs_remove_recursive(system->entry);
1018                                 list_del(&system->list);
1019                                 if (filter) {
1020                                         kfree(filter->filter_string);
1021                                         kfree(filter);
1022                                 }
1023                                 kfree(system->name);
1024                                 kfree(system);
1025                         }
1026                         break;
1027                 }
1028         }
1029 }
1030
1031 /*
1032  * Must be called under locking both of event_mutex and trace_event_mutex.
1033  */
1034 static void __trace_remove_event_call(struct ftrace_event_call *call)
1035 {
1036         ftrace_event_enable_disable(call, 0);
1037         if (call->event)
1038                 __unregister_ftrace_event(call->event);
1039         debugfs_remove_recursive(call->dir);
1040         list_del(&call->list);
1041         trace_destroy_fields(call);
1042         destroy_preds(call);
1043         remove_subsystem_dir(call->system);
1044 }
1045
1046 /* Remove an event_call */
1047 void trace_remove_event_call(struct ftrace_event_call *call)
1048 {
1049         mutex_lock(&event_mutex);
1050         down_write(&trace_event_mutex);
1051         __trace_remove_event_call(call);
1052         up_write(&trace_event_mutex);
1053         mutex_unlock(&event_mutex);
1054 }
1055
1056 #define for_each_event(event, start, end)                       \
1057         for (event = start;                                     \
1058              (unsigned long)event < (unsigned long)end;         \
1059              event++)
1060
1061 #ifdef CONFIG_MODULES
1062
1063 static LIST_HEAD(ftrace_module_file_list);
1064
1065 /*
1066  * Modules must own their file_operations to keep up with
1067  * reference counting.
1068  */
1069 struct ftrace_module_file_ops {
1070         struct list_head                list;
1071         struct module                   *mod;
1072         struct file_operations          id;
1073         struct file_operations          enable;
1074         struct file_operations          format;
1075         struct file_operations          filter;
1076 };
1077
1078 static struct ftrace_module_file_ops *
1079 trace_create_file_ops(struct module *mod)
1080 {
1081         struct ftrace_module_file_ops *file_ops;
1082
1083         /*
1084          * This is a bit of a PITA. To allow for correct reference
1085          * counting, modules must "own" their file_operations.
1086          * To do this, we allocate the file operations that will be
1087          * used in the event directory.
1088          */
1089
1090         file_ops = kmalloc(sizeof(*file_ops), GFP_KERNEL);
1091         if (!file_ops)
1092                 return NULL;
1093
1094         file_ops->mod = mod;
1095
1096         file_ops->id = ftrace_event_id_fops;
1097         file_ops->id.owner = mod;
1098
1099         file_ops->enable = ftrace_enable_fops;
1100         file_ops->enable.owner = mod;
1101
1102         file_ops->filter = ftrace_event_filter_fops;
1103         file_ops->filter.owner = mod;
1104
1105         file_ops->format = ftrace_event_format_fops;
1106         file_ops->format.owner = mod;
1107
1108         list_add(&file_ops->list, &ftrace_module_file_list);
1109
1110         return file_ops;
1111 }
1112
1113 static void trace_module_add_events(struct module *mod)
1114 {
1115         struct ftrace_module_file_ops *file_ops = NULL;
1116         struct ftrace_event_call *call, *start, *end;
1117         struct dentry *d_events;
1118         int ret;
1119
1120         start = mod->trace_events;
1121         end = mod->trace_events + mod->num_trace_events;
1122
1123         if (start == end)
1124                 return;
1125
1126         d_events = event_trace_events_dir();
1127         if (!d_events)
1128                 return;
1129
1130         for_each_event(call, start, end) {
1131                 /* The linker may leave blanks */
1132                 if (!call->name)
1133                         continue;
1134                 if (call->raw_init) {
1135                         ret = call->raw_init(call);
1136                         if (ret < 0) {
1137                                 if (ret != -ENOSYS)
1138                                         pr_warning("Could not initialize trace "
1139                                         "point events/%s\n", call->name);
1140                                 continue;
1141                         }
1142                 }
1143                 /*
1144                  * This module has events, create file ops for this module
1145                  * if not already done.
1146                  */
1147                 if (!file_ops) {
1148                         file_ops = trace_create_file_ops(mod);
1149                         if (!file_ops)
1150                                 return;
1151                 }
1152                 call->mod = mod;
1153                 ret = event_create_dir(call, d_events,
1154                                        &file_ops->id, &file_ops->enable,
1155                                        &file_ops->filter, &file_ops->format);
1156                 if (!ret)
1157                         list_add(&call->list, &ftrace_events);
1158         }
1159 }
1160
1161 static void trace_module_remove_events(struct module *mod)
1162 {
1163         struct ftrace_module_file_ops *file_ops;
1164         struct ftrace_event_call *call, *p;
1165         bool found = false;
1166
1167         down_write(&trace_event_mutex);
1168         list_for_each_entry_safe(call, p, &ftrace_events, list) {
1169                 if (call->mod == mod) {
1170                         found = true;
1171                         __trace_remove_event_call(call);
1172                 }
1173         }
1174
1175         /* Now free the file_operations */
1176         list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1177                 if (file_ops->mod == mod)
1178                         break;
1179         }
1180         if (&file_ops->list != &ftrace_module_file_list) {
1181                 list_del(&file_ops->list);
1182                 kfree(file_ops);
1183         }
1184
1185         /*
1186          * It is safest to reset the ring buffer if the module being unloaded
1187          * registered any events.
1188          */
1189         if (found)
1190                 tracing_reset_current_online_cpus();
1191         up_write(&trace_event_mutex);
1192 }
1193
1194 static int trace_module_notify(struct notifier_block *self,
1195                                unsigned long val, void *data)
1196 {
1197         struct module *mod = data;
1198
1199         mutex_lock(&event_mutex);
1200         switch (val) {
1201         case MODULE_STATE_COMING:
1202                 trace_module_add_events(mod);
1203                 break;
1204         case MODULE_STATE_GOING:
1205                 trace_module_remove_events(mod);
1206                 break;
1207         }
1208         mutex_unlock(&event_mutex);
1209
1210         return 0;
1211 }
1212 #else
1213 static int trace_module_notify(struct notifier_block *self,
1214                                unsigned long val, void *data)
1215 {
1216         return 0;
1217 }
1218 #endif /* CONFIG_MODULES */
1219
1220 static struct notifier_block trace_module_nb = {
1221         .notifier_call = trace_module_notify,
1222         .priority = 0,
1223 };
1224
1225 extern struct ftrace_event_call __start_ftrace_events[];
1226 extern struct ftrace_event_call __stop_ftrace_events[];
1227
1228 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
1229
1230 static __init int setup_trace_event(char *str)
1231 {
1232         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
1233         ring_buffer_expanded = 1;
1234         tracing_selftest_disabled = 1;
1235
1236         return 1;
1237 }
1238 __setup("trace_event=", setup_trace_event);
1239
1240 static __init int event_trace_init(void)
1241 {
1242         struct ftrace_event_call *call;
1243         struct dentry *d_tracer;
1244         struct dentry *entry;
1245         struct dentry *d_events;
1246         int ret;
1247         char *buf = bootup_event_buf;
1248         char *token;
1249
1250         d_tracer = tracing_init_dentry();
1251         if (!d_tracer)
1252                 return 0;
1253
1254         entry = debugfs_create_file("available_events", 0444, d_tracer,
1255                                     (void *)&show_event_seq_ops,
1256                                     &ftrace_avail_fops);
1257         if (!entry)
1258                 pr_warning("Could not create debugfs "
1259                            "'available_events' entry\n");
1260
1261         entry = debugfs_create_file("set_event", 0644, d_tracer,
1262                                     (void *)&show_set_event_seq_ops,
1263                                     &ftrace_set_event_fops);
1264         if (!entry)
1265                 pr_warning("Could not create debugfs "
1266                            "'set_event' entry\n");
1267
1268         d_events = event_trace_events_dir();
1269         if (!d_events)
1270                 return 0;
1271
1272         /* ring buffer internal formats */
1273         trace_create_file("header_page", 0444, d_events,
1274                           ring_buffer_print_page_header,
1275                           &ftrace_show_header_fops);
1276
1277         trace_create_file("header_event", 0444, d_events,
1278                           ring_buffer_print_entry_header,
1279                           &ftrace_show_header_fops);
1280
1281         trace_create_file("enable", 0644, d_events,
1282                           NULL, &ftrace_system_enable_fops);
1283
1284         for_each_event(call, __start_ftrace_events, __stop_ftrace_events) {
1285                 /* The linker may leave blanks */
1286                 if (!call->name)
1287                         continue;
1288                 if (call->raw_init) {
1289                         ret = call->raw_init(call);
1290                         if (ret < 0) {
1291                                 if (ret != -ENOSYS)
1292                                         pr_warning("Could not initialize trace "
1293                                         "point events/%s\n", call->name);
1294                                 continue;
1295                         }
1296                 }
1297                 ret = event_create_dir(call, d_events, &ftrace_event_id_fops,
1298                                        &ftrace_enable_fops,
1299                                        &ftrace_event_filter_fops,
1300                                        &ftrace_event_format_fops);
1301                 if (!ret)
1302                         list_add(&call->list, &ftrace_events);
1303         }
1304
1305         while (true) {
1306                 token = strsep(&buf, ",");
1307
1308                 if (!token)
1309                         break;
1310                 if (!*token)
1311                         continue;
1312
1313                 ret = ftrace_set_clr_event(token, 1);
1314                 if (ret)
1315                         pr_warning("Failed to enable trace event: %s\n", token);
1316         }
1317
1318         ret = register_module_notifier(&trace_module_nb);
1319         if (ret)
1320                 pr_warning("Failed to register trace events module notifier\n");
1321
1322         return 0;
1323 }
1324 fs_initcall(event_trace_init);
1325
1326 #ifdef CONFIG_FTRACE_STARTUP_TEST
1327
1328 static DEFINE_SPINLOCK(test_spinlock);
1329 static DEFINE_SPINLOCK(test_spinlock_irq);
1330 static DEFINE_MUTEX(test_mutex);
1331
1332 static __init void test_work(struct work_struct *dummy)
1333 {
1334         spin_lock(&test_spinlock);
1335         spin_lock_irq(&test_spinlock_irq);
1336         udelay(1);
1337         spin_unlock_irq(&test_spinlock_irq);
1338         spin_unlock(&test_spinlock);
1339
1340         mutex_lock(&test_mutex);
1341         msleep(1);
1342         mutex_unlock(&test_mutex);
1343 }
1344
1345 static __init int event_test_thread(void *unused)
1346 {
1347         void *test_malloc;
1348
1349         test_malloc = kmalloc(1234, GFP_KERNEL);
1350         if (!test_malloc)
1351                 pr_info("failed to kmalloc\n");
1352
1353         schedule_on_each_cpu(test_work);
1354
1355         kfree(test_malloc);
1356
1357         set_current_state(TASK_INTERRUPTIBLE);
1358         while (!kthread_should_stop())
1359                 schedule();
1360
1361         return 0;
1362 }
1363
1364 /*
1365  * Do various things that may trigger events.
1366  */
1367 static __init void event_test_stuff(void)
1368 {
1369         struct task_struct *test_thread;
1370
1371         test_thread = kthread_run(event_test_thread, NULL, "test-events");
1372         msleep(1);
1373         kthread_stop(test_thread);
1374 }
1375
1376 /*
1377  * For every trace event defined, we will test each trace point separately,
1378  * and then by groups, and finally all trace points.
1379  */
1380 static __init void event_trace_self_tests(void)
1381 {
1382         struct ftrace_event_call *call;
1383         struct event_subsystem *system;
1384         int ret;
1385
1386         pr_info("Running tests on trace events:\n");
1387
1388         list_for_each_entry(call, &ftrace_events, list) {
1389
1390                 /* Only test those that have a regfunc */
1391                 if (!call->regfunc)
1392                         continue;
1393
1394 /*
1395  * Testing syscall events here is pretty useless, but
1396  * we still do it if configured. But this is time consuming.
1397  * What we really need is a user thread to perform the
1398  * syscalls as we test.
1399  */
1400 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
1401                 if (call->system &&
1402                     strcmp(call->system, "syscalls") == 0)
1403                         continue;
1404 #endif
1405
1406                 pr_info("Testing event %s: ", call->name);
1407
1408                 /*
1409                  * If an event is already enabled, someone is using
1410                  * it and the self test should not be on.
1411                  */
1412                 if (call->enabled) {
1413                         pr_warning("Enabled event during self test!\n");
1414                         WARN_ON_ONCE(1);
1415                         continue;
1416                 }
1417
1418                 ftrace_event_enable_disable(call, 1);
1419                 event_test_stuff();
1420                 ftrace_event_enable_disable(call, 0);
1421
1422                 pr_cont("OK\n");
1423         }
1424
1425         /* Now test at the sub system level */
1426
1427         pr_info("Running tests on trace event systems:\n");
1428
1429         list_for_each_entry(system, &event_subsystems, list) {
1430
1431                 /* the ftrace system is special, skip it */
1432                 if (strcmp(system->name, "ftrace") == 0)
1433                         continue;
1434
1435                 pr_info("Testing event system %s: ", system->name);
1436
1437                 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 1);
1438                 if (WARN_ON_ONCE(ret)) {
1439                         pr_warning("error enabling system %s\n",
1440                                    system->name);
1441                         continue;
1442                 }
1443
1444                 event_test_stuff();
1445
1446                 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 0);
1447                 if (WARN_ON_ONCE(ret))
1448                         pr_warning("error disabling system %s\n",
1449                                    system->name);
1450
1451                 pr_cont("OK\n");
1452         }
1453
1454         /* Test with all events enabled */
1455
1456         pr_info("Running tests on all trace events:\n");
1457         pr_info("Testing all events: ");
1458
1459         ret = __ftrace_set_clr_event(NULL, NULL, NULL, 1);
1460         if (WARN_ON_ONCE(ret)) {
1461                 pr_warning("error enabling all events\n");
1462                 return;
1463         }
1464
1465         event_test_stuff();
1466
1467         /* reset sysname */
1468         ret = __ftrace_set_clr_event(NULL, NULL, NULL, 0);
1469         if (WARN_ON_ONCE(ret)) {
1470                 pr_warning("error disabling all events\n");
1471                 return;
1472         }
1473
1474         pr_cont("OK\n");
1475 }
1476
1477 #ifdef CONFIG_FUNCTION_TRACER
1478
1479 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
1480
1481 static void
1482 function_test_events_call(unsigned long ip, unsigned long parent_ip)
1483 {
1484         struct ring_buffer_event *event;
1485         struct ring_buffer *buffer;
1486         struct ftrace_entry *entry;
1487         unsigned long flags;
1488         long disabled;
1489         int resched;
1490         int cpu;
1491         int pc;
1492
1493         pc = preempt_count();
1494         resched = ftrace_preempt_disable();
1495         cpu = raw_smp_processor_id();
1496         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
1497
1498         if (disabled != 1)
1499                 goto out;
1500
1501         local_save_flags(flags);
1502
1503         event = trace_current_buffer_lock_reserve(&buffer,
1504                                                   TRACE_FN, sizeof(*entry),
1505                                                   flags, pc);
1506         if (!event)
1507                 goto out;
1508         entry   = ring_buffer_event_data(event);
1509         entry->ip                       = ip;
1510         entry->parent_ip                = parent_ip;
1511
1512         trace_nowake_buffer_unlock_commit(buffer, event, flags, pc);
1513
1514  out:
1515         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
1516         ftrace_preempt_enable(resched);
1517 }
1518
1519 static struct ftrace_ops trace_ops __initdata  =
1520 {
1521         .func = function_test_events_call,
1522 };
1523
1524 static __init void event_trace_self_test_with_function(void)
1525 {
1526         register_ftrace_function(&trace_ops);
1527         pr_info("Running tests again, along with the function tracer\n");
1528         event_trace_self_tests();
1529         unregister_ftrace_function(&trace_ops);
1530 }
1531 #else
1532 static __init void event_trace_self_test_with_function(void)
1533 {
1534 }
1535 #endif
1536
1537 static __init int event_trace_self_tests_init(void)
1538 {
1539         if (!tracing_selftest_disabled) {
1540                 event_trace_self_tests();
1541                 event_trace_self_test_with_function();
1542         }
1543
1544         return 0;
1545 }
1546
1547 late_initcall(event_trace_self_tests_init);
1548
1549 #endif