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/slab.h>
  19 #include <linux/delay.h>
  20
  21 #include <asm/setup.h>
  22
  23 #include "trace_output.h"
  24
  25 #undef TRACE_SYSTEM
  26 #define TRACE_SYSTEM "TRACE_SYSTEM"
  27
  28 DEFINE_MUTEX(event_mutex);
  29
  30 LIST_HEAD(ftrace_events);
  31
  32 struct list_head *
  33 trace_get_fields(struct ftrace_event_call *event_call)
  34 {
  35         if (!event_call->class->get_fields)
  36                 return &event_call->class->fields;
  37         return event_call->class->get_fields(event_call);
  38 }
  39
  40 int trace_define_field(struct ftrace_event_call *call, const char *type,
  41                        const char *name, int offset, int size, int is_signed,
  42                        int filter_type)
  43 {
  44         struct ftrace_event_field *field;
  45         struct list_head *head;
  46
  47         if (WARN_ON(!call->class))
  48                 return 0;
  49
  50         field = kzalloc(sizeof(*field), GFP_KERNEL);
  51         if (!field)
  52                 goto err;
  53
  54         field->name = kstrdup(name, GFP_KERNEL);
  55         if (!field->name)
  56                 goto err;
  57
  58         field->type = kstrdup(type, GFP_KERNEL);
  59         if (!field->type)
  60                 goto err;
  61
  62         if (filter_type == FILTER_OTHER)
  63                 field->filter_type = filter_assign_type(type);
  64         else
  65                 field->filter_type = filter_type;
  66
  67         field->offset = offset;
  68         field->size = size;
  69         field->is_signed = is_signed;
  70
  71         head = trace_get_fields(call);
  72         list_add(&field->link, head);
  73
  74         return 0;
  75
  76 err:
  77         if (field)
  78                 kfree(field->name);
  79         kfree(field);
  80
  81         return -ENOMEM;
  82 }
  83 EXPORT_SYMBOL_GPL(trace_define_field);
  84
  85 #define __common_field(type, item)                                      \
  86         ret = trace_define_field(call, #type, "common_" #item,          \
  87                                  offsetof(typeof(ent), item),           \
  88                                  sizeof(ent.item),                      \
  89                                  is_signed_type(type), FILTER_OTHER);   \
  90         if (ret)                                                        \
  91                 return ret;
  92
  93 static int trace_define_common_fields(struct ftrace_event_call *call)
  94 {
  95         int ret;
  96         struct trace_entry ent;
  97
  98         __common_field(unsigned short, type);
  99         __common_field(unsigned char, flags);
 100         __common_field(unsigned char, preempt_count);
 101         __common_field(int, pid);
 102         __common_field(int, lock_depth);
 103
 104         return ret;
 105 }
 106
 107 void trace_destroy_fields(struct ftrace_event_call *call)
 108 {
 109         struct ftrace_event_field *field, *next;
 110         struct list_head *head;
 111
 112         head = trace_get_fields(call);
 113         list_for_each_entry_safe(field, next, head, link) {
 114                 list_del(&field->link);
 115                 kfree(field->type);
 116                 kfree(field->name);
 117                 kfree(field);
 118         }
 119 }
 120
 121 int trace_event_raw_init(struct ftrace_event_call *call)
 122 {
 123         int id;
 124
 125         id = register_ftrace_event(&call->event);
 126         if (!id)
 127                 return -ENODEV;
 128
 129         return 0;
 130 }
 131 EXPORT_SYMBOL_GPL(trace_event_raw_init);
 132
 133 static int ftrace_event_enable_disable(struct ftrace_event_call *call,
 134                                         int enable)
 135 {
 136         int ret = 0;
 137
 138         switch (enable) {
 139         case 0:
 140                 if (call->flags & TRACE_EVENT_FL_ENABLED) {
 141                         call->flags &= ~TRACE_EVENT_FL_ENABLED;
 142                         tracing_stop_cmdline_record();
 143                         if (call->class->reg)
 144                                 call->class->reg(call, TRACE_REG_UNREGISTER);
 145                         else
 146                                 tracepoint_probe_unregister(call->name,
 147                                                             call->class->probe,
 148                                                             call);
 149                 }
 150                 break;
 151         case 1:
 152                 if (!(call->flags & TRACE_EVENT_FL_ENABLED)) {
 153                         tracing_start_cmdline_record();
 154                         if (call->class->reg)
 155                                 ret = call->class->reg(call, TRACE_REG_REGISTER);
 156                         else
 157                                 ret = tracepoint_probe_register(call->name,
 158                                                                 call->class->probe,
 159                                                                 call);
 160                         if (ret) {
 161                                 tracing_stop_cmdline_record();
 162                                 pr_info("event trace: Could not enable event "
 163                                         "%s\n", call->name);
 164                                 break;
 165                         }
 166                         call->flags |= TRACE_EVENT_FL_ENABLED;
 167                 }
 168                 break;
 169         }
 170
 171         return ret;
 172 }
 173
 174 static void ftrace_clear_events(void)
 175 {
 176         struct ftrace_event_call *call;
 177
 178         mutex_lock(&event_mutex);
 179         list_for_each_entry(call, &ftrace_events, list) {
 180                 ftrace_event_enable_disable(call, 0);
 181         }
 182         mutex_unlock(&event_mutex);
 183 }
 184
 185 /*
 186  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 187  */
 188 static int __ftrace_set_clr_event(const char *match, const char *sub,
 189                                   const char *event, int set)
 190 {
 191         struct ftrace_event_call *call;
 192         int ret = -EINVAL;
 193
 194         mutex_lock(&event_mutex);
 195         list_for_each_entry(call, &ftrace_events, list) {
 196
 197                 if (!call->name || !call->class ||
 198                     (!call->class->probe && !call->class->reg))
 199                         continue;
 200
 201                 if (match &&
 202                     strcmp(match, call->name) != 0 &&
 203                     strcmp(match, call->class->system) != 0)
 204                         continue;
 205
 206                 if (sub && strcmp(sub, call->class->system) != 0)
 207                         continue;
 208
 209                 if (event && strcmp(event, call->name) != 0)
 210                         continue;
 211
 212                 ftrace_event_enable_disable(call, set);
 213
 214                 ret = 0;
 215         }
 216         mutex_unlock(&event_mutex);
 217
 218         return ret;
 219 }
 220
 221 static int ftrace_set_clr_event(char *buf, int set)
 222 {
 223         char *event = NULL, *sub = NULL, *match;
 224
 225         /*
 226          * The buf format can be <subsystem>:<event-name>
 227          *  *:<event-name> means any event by that name.
 228          *  :<event-name> is the same.
 229          *
 230          *  <subsystem>:* means all events in that subsystem
 231          *  <subsystem>: means the same.
 232          *
 233          *  <name> (no ':') means all events in a subsystem with
 234          *  the name <name> or any event that matches <name>
 235          */
 236
 237         match = strsep(&buf, ":");
 238         if (buf) {
 239                 sub = match;
 240                 event = buf;
 241                 match = NULL;
 242
 243                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 244                         sub = NULL;
 245                 if (!strlen(event) || strcmp(event, "*") == 0)
 246                         event = NULL;
 247         }
 248
 249         return __ftrace_set_clr_event(match, sub, event, set);
 250 }
 251
 252 /**
 253  * trace_set_clr_event - enable or disable an event
 254  * @system: system name to match (NULL for any system)
 255  * @event: event name to match (NULL for all events, within system)
 256  * @set: 1 to enable, 0 to disable
 257  *
 258  * This is a way for other parts of the kernel to enable or disable
 259  * event recording.
 260  *
 261  * Returns 0 on success, -EINVAL if the parameters do not match any
 262  * registered events.
 263  */
 264 int trace_set_clr_event(const char *system, const char *event, int set)
 265 {
 266         return __ftrace_set_clr_event(NULL, system, event, set);
 267 }
 268
 269 /* 128 should be much more than enough */
 270 #define EVENT_BUF_SIZE          127
 271
 272 static ssize_t
 273 ftrace_event_write(struct file *file, const char __user *ubuf,
 274                    size_t cnt, loff_t *ppos)
 275 {
 276         struct trace_parser parser;
 277         ssize_t read, ret;
 278
 279         if (!cnt)
 280                 return 0;
 281
 282         ret = tracing_update_buffers();
 283         if (ret < 0)
 284                 return ret;
 285
 286         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 287                 return -ENOMEM;
 288
 289         read = trace_get_user(&parser, ubuf, cnt, ppos);
 290
 291         if (read >= 0 && trace_parser_loaded((&parser))) {
 292                 int set = 1;
 293
 294                 if (*parser.buffer == '!')
 295                         set = 0;
 296
 297                 parser.buffer[parser.idx] = 0;
 298
 299                 ret = ftrace_set_clr_event(parser.buffer + !set, set);
 300                 if (ret)
 301                         goto out_put;
 302         }
 303
 304         ret = read;
 305
 306  out_put:
 307         trace_parser_put(&parser);
 308
 309         return ret;
 310 }
 311
 312 static void *
 313 t_next(struct seq_file *m, void *v, loff_t *pos)
 314 {
 315         struct ftrace_event_call *call = v;
 316
 317         (*pos)++;
 318
 319         list_for_each_entry_continue(call, &ftrace_events, list) {
 320                 /*
 321                  * The ftrace subsystem is for showing formats only.
 322                  * They can not be enabled or disabled via the event files.
 323                  */
 324                 if (call->class && (call->class->probe || call->class->reg))
 325                         return call;
 326         }
 327
 328         return NULL;
 329 }
 330
 331 static void *t_start(struct seq_file *m, loff_t *pos)
 332 {
 333         struct ftrace_event_call *call;
 334         loff_t l;
 335
 336         mutex_lock(&event_mutex);
 337
 338         call = list_entry(&ftrace_events, struct ftrace_event_call, list);
 339         for (l = 0; l <= *pos; ) {
 340                 call = t_next(m, call, &l);
 341                 if (!call)
 342                         break;
 343         }
 344         return call;
 345 }
 346
 347 static void *
 348 s_next(struct seq_file *m, void *v, loff_t *pos)
 349 {
 350         struct ftrace_event_call *call = v;
 351
 352         (*pos)++;
 353
 354         list_for_each_entry_continue(call, &ftrace_events, list) {
 355                 if (call->flags & TRACE_EVENT_FL_ENABLED)
 356                         return call;
 357         }
 358
 359         return NULL;
 360 }
 361
 362 static void *s_start(struct seq_file *m, loff_t *pos)
 363 {
 364         struct ftrace_event_call *call;
 365         loff_t l;
 366
 367         mutex_lock(&event_mutex);
 368
 369         call = list_entry(&ftrace_events, struct ftrace_event_call, list);
 370         for (l = 0; l <= *pos; ) {
 371                 call = s_next(m, call, &l);
 372                 if (!call)
 373                         break;
 374         }
 375         return call;
 376 }
 377
 378 static int t_show(struct seq_file *m, void *v)
 379 {
 380         struct ftrace_event_call *call = v;
 381
 382         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
 383                 seq_printf(m, "%s:", call->class->system);
 384         seq_printf(m, "%s\n", call->name);
 385
 386         return 0;
 387 }
 388
 389 static void t_stop(struct seq_file *m, void *p)
 390 {
 391         mutex_unlock(&event_mutex);
 392 }
 393
 394 static int
 395 ftrace_event_seq_open(struct inode *inode, struct file *file)
 396 {
 397         const struct seq_operations *seq_ops;
 398
 399         if ((file->f_mode & FMODE_WRITE) &&
 400             (file->f_flags & O_TRUNC))
 401                 ftrace_clear_events();
 402
 403         seq_ops = inode->i_private;
 404         return seq_open(file, seq_ops);
 405 }
 406
 407 static ssize_t
 408 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 409                   loff_t *ppos)
 410 {
 411         struct ftrace_event_call *call = filp->private_data;
 412         char *buf;
 413
 414         if (call->flags & TRACE_EVENT_FL_ENABLED)
 415                 buf = "1\n";
 416         else
 417                 buf = "0\n";
 418
 419         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 420 }
 421
 422 static ssize_t
 423 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 424                    loff_t *ppos)
 425 {
 426         struct ftrace_event_call *call = filp->private_data;
 427         char buf[64];
 428         unsigned long val;
 429         int ret;
 430
 431         if (cnt >= sizeof(buf))
 432                 return -EINVAL;
 433
 434         if (copy_from_user(&buf, ubuf, cnt))
 435                 return -EFAULT;
 436
 437         buf[cnt] = 0;
 438
 439         ret = strict_strtoul(buf, 10, &val);
 440         if (ret < 0)
 441                 return ret;
 442
 443         ret = tracing_update_buffers();
 444         if (ret < 0)
 445                 return ret;
 446
 447         switch (val) {
 448         case 0:
 449         case 1:
 450                 mutex_lock(&event_mutex);
 451                 ret = ftrace_event_enable_disable(call, val);
 452                 mutex_unlock(&event_mutex);
 453                 break;
 454
 455         default:
 456                 return -EINVAL;
 457         }
 458
 459         *ppos += cnt;
 460
 461         return ret ? ret : cnt;
 462 }
 463
 464 static ssize_t
 465 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 466                    loff_t *ppos)
 467 {
 468         const char set_to_char[4] = { '?', '0', '1', 'X' };
 469         const char *system = filp->private_data;
 470         struct ftrace_event_call *call;
 471         char buf[2];
 472         int set = 0;
 473         int ret;
 474
 475         mutex_lock(&event_mutex);
 476         list_for_each_entry(call, &ftrace_events, list) {
 477                 if (!call->name || !call->class ||
 478                     (!call->class->probe && !call->class->reg))
 479                         continue;
 480
 481                 if (system && strcmp(call->class->system, system) != 0)
 482                         continue;
 483
 484                 /*
 485                  * We need to find out if all the events are set
 486                  * or if all events or cleared, or if we have
 487                  * a mixture.
 488                  */
 489                 set |= (1 << !!(call->flags & TRACE_EVENT_FL_ENABLED));
 490
 491                 /*
 492                  * If we have a mixture, no need to look further.
 493                  */
 494                 if (set == 3)
 495                         break;
 496         }
 497         mutex_unlock(&event_mutex);
 498
 499         buf[0] = set_to_char[set];
 500         buf[1] = '\n';
 501
 502         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 503
 504         return ret;
 505 }
 506
 507 static ssize_t
 508 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 509                     loff_t *ppos)
 510 {
 511         const char *system = filp->private_data;
 512         unsigned long val;
 513         char buf[64];
 514         ssize_t ret;
 515
 516         if (cnt >= sizeof(buf))
 517                 return -EINVAL;
 518
 519         if (copy_from_user(&buf, ubuf, cnt))
 520                 return -EFAULT;
 521
 522         buf[cnt] = 0;
 523
 524         ret = strict_strtoul(buf, 10, &val);
 525         if (ret < 0)
 526                 return ret;
 527
 528         ret = tracing_update_buffers();
 529         if (ret < 0)
 530                 return ret;
 531
 532         if (val != 0 && val != 1)
 533                 return -EINVAL;
 534
 535         ret = __ftrace_set_clr_event(NULL, system, NULL, val);
 536         if (ret)
 537                 goto out;
 538
 539         ret = cnt;
 540
 541 out:
 542         *ppos += cnt;
 543
 544         return ret;
 545 }
 546
 547 static ssize_t
 548 event_format_read(struct file *filp, char __user *ubuf, size_t cnt,
 549                   loff_t *ppos)
 550 {
 551         struct ftrace_event_call *call = filp->private_data;
 552         struct ftrace_event_field *field;
 553         struct list_head *head;
 554         struct trace_seq *s;
 555         int common_field_count = 5;
 556         char *buf;
 557         int r = 0;
 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         trace_seq_printf(s, "name: %s\n", call->name);
 569         trace_seq_printf(s, "ID: %d\n", call->event.type);
 570         trace_seq_printf(s, "format:\n");
 571
 572         head = trace_get_fields(call);
 573         list_for_each_entry_reverse(field, head, link) {
 574                 /*
 575                  * Smartly shows the array type(except dynamic array).
 576                  * Normal:
 577                  *      field:TYPE VAR
 578                  * If TYPE := TYPE[LEN], it is shown:
 579                  *      field:TYPE VAR[LEN]
 580                  */
 581                 const char *array_descriptor = strchr(field->type, '[');
 582
 583                 if (!strncmp(field->type, "__data_loc", 10))
 584                         array_descriptor = NULL;
 585
 586                 if (!array_descriptor) {
 587                         r = trace_seq_printf(s, "\tfield:%s %s;\toffset:%u;"
 588                                         "\tsize:%u;\tsigned:%d;\n",
 589                                         field->type, field->name, field->offset,
 590                                         field->size, !!field->is_signed);
 591                 } else {
 592                         r = trace_seq_printf(s, "\tfield:%.*s %s%s;\toffset:%u;"
 593                                         "\tsize:%u;\tsigned:%d;\n",
 594                                         (int)(array_descriptor - field->type),
 595                                         field->type, field->name,
 596                                         array_descriptor, field->offset,
 597                                         field->size, !!field->is_signed);
 598                 }
 599
 600                 if (--common_field_count == 0)
 601                         r = trace_seq_printf(s, "\n");
 602
 603                 if (!r)
 604                         break;
 605         }
 606
 607         if (r)
 608                 r = trace_seq_printf(s, "\nprint fmt: %s\n",
 609                                 call->print_fmt);
 610
 611         if (!r) {
 612                 /*
 613                  * ug!  The format output is bigger than a PAGE!!
 614                  */
 615                 buf = "FORMAT TOO BIG\n";
 616                 r = simple_read_from_buffer(ubuf, cnt, ppos,
 617                                               buf, strlen(buf));
 618                 goto out;
 619         }
 620
 621         r = simple_read_from_buffer(ubuf, cnt, ppos,
 622                                     s->buffer, s->len);
 623  out:
 624         kfree(s);
 625         return r;
 626 }
 627
 628 static ssize_t
 629 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 630 {
 631         struct ftrace_event_call *call = filp->private_data;
 632         struct trace_seq *s;
 633         int r;
 634
 635         if (*ppos)
 636                 return 0;
 637
 638         s = kmalloc(sizeof(*s), GFP_KERNEL);
 639         if (!s)
 640                 return -ENOMEM;
 641
 642         trace_seq_init(s);
 643         trace_seq_printf(s, "%d\n", call->event.type);
 644
 645         r = simple_read_from_buffer(ubuf, cnt, ppos,
 646                                     s->buffer, s->len);
 647         kfree(s);
 648         return r;
 649 }
 650
 651 static ssize_t
 652 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 653                   loff_t *ppos)
 654 {
 655         struct ftrace_event_call *call = filp->private_data;
 656         struct trace_seq *s;
 657         int r;
 658
 659         if (*ppos)
 660                 return 0;
 661
 662         s = kmalloc(sizeof(*s), GFP_KERNEL);
 663         if (!s)
 664                 return -ENOMEM;
 665
 666         trace_seq_init(s);
 667
 668         print_event_filter(call, s);
 669         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 670
 671         kfree(s);
 672
 673         return r;
 674 }
 675
 676 static ssize_t
 677 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
 678                    loff_t *ppos)
 679 {
 680         struct ftrace_event_call *call = filp->private_data;
 681         char *buf;
 682         int err;
 683
 684         if (cnt >= PAGE_SIZE)
 685                 return -EINVAL;
 686
 687         buf = (char *)__get_free_page(GFP_TEMPORARY);
 688         if (!buf)
 689                 return -ENOMEM;
 690
 691         if (copy_from_user(buf, ubuf, cnt)) {
 692                 free_page((unsigned long) buf);
 693                 return -EFAULT;
 694         }
 695         buf[cnt] = '\0';
 696
 697         err = apply_event_filter(call, buf);
 698         free_page((unsigned long) buf);
 699         if (err < 0)
 700                 return err;
 701
 702         *ppos += cnt;
 703
 704         return cnt;
 705 }
 706
 707 static ssize_t
 708 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 709                       loff_t *ppos)
 710 {
 711         struct event_subsystem *system = filp->private_data;
 712         struct trace_seq *s;
 713         int r;
 714
 715         if (*ppos)
 716                 return 0;
 717
 718         s = kmalloc(sizeof(*s), GFP_KERNEL);
 719         if (!s)
 720                 return -ENOMEM;
 721
 722         trace_seq_init(s);
 723
 724         print_subsystem_event_filter(system, s);
 725         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 726
 727         kfree(s);
 728
 729         return r;
 730 }
 731
 732 static ssize_t
 733 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
 734                        loff_t *ppos)
 735 {
 736         struct event_subsystem *system = filp->private_data;
 737         char *buf;
 738         int err;
 739
 740         if (cnt >= PAGE_SIZE)
 741                 return -EINVAL;
 742
 743         buf = (char *)__get_free_page(GFP_TEMPORARY);
 744         if (!buf)
 745                 return -ENOMEM;
 746
 747         if (copy_from_user(buf, ubuf, cnt)) {
 748                 free_page((unsigned long) buf);
 749                 return -EFAULT;
 750         }
 751         buf[cnt] = '\0';
 752
 753         err = apply_subsystem_event_filter(system, buf);
 754         free_page((unsigned long) buf);
 755         if (err < 0)
 756                 return err;
 757
 758         *ppos += cnt;
 759
 760         return cnt;
 761 }
 762
 763 static ssize_t
 764 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 765 {
 766         int (*func)(struct trace_seq *s) = filp->private_data;
 767         struct trace_seq *s;
 768         int r;
 769
 770         if (*ppos)
 771                 return 0;
 772
 773         s = kmalloc(sizeof(*s), GFP_KERNEL);
 774         if (!s)
 775                 return -ENOMEM;
 776
 777         trace_seq_init(s);
 778
 779         func(s);
 780         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 781
 782         kfree(s);
 783
 784         return r;
 785 }
 786
 787 static const struct seq_operations show_event_seq_ops = {
 788         .start = t_start,
 789         .next = t_next,
 790         .show = t_show,
 791         .stop = t_stop,
 792 };
 793
 794 static const struct seq_operations show_set_event_seq_ops = {
 795         .start = s_start,
 796         .next = s_next,
 797         .show = t_show,
 798         .stop = t_stop,
 799 };
 800
 801 static const struct file_operations ftrace_avail_fops = {
 802         .open = ftrace_event_seq_open,
 803         .read = seq_read,
 804         .llseek = seq_lseek,
 805         .release = seq_release,
 806 };
 807
 808 static const struct file_operations ftrace_set_event_fops = {
 809         .open = ftrace_event_seq_open,
 810         .read = seq_read,
 811         .write = ftrace_event_write,
 812         .llseek = seq_lseek,
 813         .release = seq_release,
 814 };
 815
 816 static const struct file_operations ftrace_enable_fops = {
 817         .open = tracing_open_generic,
 818         .read = event_enable_read,
 819         .write = event_enable_write,
 820 };
 821
 822 static const struct file_operations ftrace_event_format_fops = {
 823         .open = tracing_open_generic,
 824         .read = event_format_read,
 825 };
 826
 827 static const struct file_operations ftrace_event_id_fops = {
 828         .open = tracing_open_generic,
 829         .read = event_id_read,
 830 };
 831
 832 static const struct file_operations ftrace_event_filter_fops = {
 833         .open = tracing_open_generic,
 834         .read = event_filter_read,
 835         .write = event_filter_write,
 836 };
 837
 838 static const struct file_operations ftrace_subsystem_filter_fops = {
 839         .open = tracing_open_generic,
 840         .read = subsystem_filter_read,
 841         .write = subsystem_filter_write,
 842 };
 843
 844 static const struct file_operations ftrace_system_enable_fops = {
 845         .open = tracing_open_generic,
 846         .read = system_enable_read,
 847         .write = system_enable_write,
 848 };
 849
 850 static const struct file_operations ftrace_show_header_fops = {
 851         .open = tracing_open_generic,
 852         .read = show_header,
 853 };
 854
 855 static struct dentry *event_trace_events_dir(void)
 856 {
 857         static struct dentry *d_tracer;
 858         static struct dentry *d_events;
 859
 860         if (d_events)
 861                 return d_events;
 862
 863         d_tracer = tracing_init_dentry();
 864         if (!d_tracer)
 865                 return NULL;
 866
 867         d_events = debugfs_create_dir("events", d_tracer);
 868         if (!d_events)
 869                 pr_warning("Could not create debugfs "
 870                            "'events' directory\n");
 871
 872         return d_events;
 873 }
 874
 875 static LIST_HEAD(event_subsystems);
 876
 877 static struct dentry *
 878 event_subsystem_dir(const char *name, struct dentry *d_events)
 879 {
 880         struct event_subsystem *system;
 881         struct dentry *entry;
 882
 883         /* First see if we did not already create this dir */
 884         list_for_each_entry(system, &event_subsystems, list) {
 885                 if (strcmp(system->name, name) == 0) {
 886                         system->nr_events++;
 887                         return system->entry;
 888                 }
 889         }
 890
 891         /* need to create new entry */
 892         system = kmalloc(sizeof(*system), GFP_KERNEL);
 893         if (!system) {
 894                 pr_warning("No memory to create event subsystem %s\n",
 895                            name);
 896                 return d_events;
 897         }
 898
 899         system->entry = debugfs_create_dir(name, d_events);
 900         if (!system->entry) {
 901                 pr_warning("Could not create event subsystem %s\n",
 902                            name);
 903                 kfree(system);
 904                 return d_events;
 905         }
 906
 907         system->nr_events = 1;
 908         system->name = kstrdup(name, GFP_KERNEL);
 909         if (!system->name) {
 910                 debugfs_remove(system->entry);
 911                 kfree(system);
 912                 return d_events;
 913         }
 914
 915         list_add(&system->list, &event_subsystems);
 916
 917         system->filter = NULL;
 918
 919         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
 920         if (!system->filter) {
 921                 pr_warning("Could not allocate filter for subsystem "
 922                            "'%s'\n", name);
 923                 return system->entry;
 924         }
 925
 926         entry = debugfs_create_file("filter", 0644, system->entry, system,
 927                                     &ftrace_subsystem_filter_fops);
 928         if (!entry) {
 929                 kfree(system->filter);
 930                 system->filter = NULL;
 931                 pr_warning("Could not create debugfs "
 932                            "'%s/filter' entry\n", name);
 933         }
 934
 935         trace_create_file("enable", 0644, system->entry,
 936                           (void *)system->name,
 937                           &ftrace_system_enable_fops);
 938
 939         return system->entry;
 940 }
 941
 942 static int
 943 event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
 944                  const struct file_operations *id,
 945                  const struct file_operations *enable,
 946                  const struct file_operations *filter,
 947                  const struct file_operations *format)
 948 {
 949         struct list_head *head;
 950         int ret;
 951
 952         /*
 953          * If the trace point header did not define TRACE_SYSTEM
 954          * then the system would be called "TRACE_SYSTEM".
 955          */
 956         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
 957                 d_events = event_subsystem_dir(call->class->system, d_events);
 958
 959         call->dir = debugfs_create_dir(call->name, d_events);
 960         if (!call->dir) {
 961                 pr_warning("Could not create debugfs "
 962                            "'%s' directory\n", call->name);
 963                 return -1;
 964         }
 965
 966         if (call->class->probe || call->class->reg)
 967                 trace_create_file("enable", 0644, call->dir, call,
 968                                   enable);
 969
 970 #ifdef CONFIG_PERF_EVENTS
 971         if (call->event.type && (call->class->perf_probe || call->class->reg))
 972                 trace_create_file("id", 0444, call->dir, call,
 973                                   id);
 974 #endif
 975
 976         if (call->class->define_fields) {
 977                 /*
 978                  * Other events may have the same class. Only update
 979                  * the fields if they are not already defined.
 980                  */
 981                 head = trace_get_fields(call);
 982                 if (list_empty(head)) {
 983                         ret = trace_define_common_fields(call);
 984                         if (!ret)
 985                                 ret = call->class->define_fields(call);
 986                         if (ret < 0) {
 987                                 pr_warning("Could not initialize trace point"
 988                                            " events/%s\n", call->name);
 989                                 return ret;
 990                         }
 991                 }
 992                 trace_create_file("filter", 0644, call->dir, call,
 993                                   filter);
 994         }
 995
 996         trace_create_file("format", 0444, call->dir, call,
 997                           format);
 998
 999         return 0;
1000 }
1001
1002 static int __trace_add_event_call(struct ftrace_event_call *call)
1003 {
1004         struct dentry *d_events;
1005         int ret;
1006
1007         if (!call->name)
1008                 return -EINVAL;
1009
1010         if (call->class->raw_init) {
1011                 ret = call->class->raw_init(call);
1012                 if (ret < 0) {
1013                         if (ret != -ENOSYS)
1014                                 pr_warning("Could not initialize trace "
1015                                 "events/%s\n", call->name);
1016                         return ret;
1017                 }
1018         }
1019
1020         d_events = event_trace_events_dir();
1021         if (!d_events)
1022                 return -ENOENT;
1023
1024         ret = event_create_dir(call, d_events, &ftrace_event_id_fops,
1025                                 &ftrace_enable_fops, &ftrace_event_filter_fops,
1026                                 &ftrace_event_format_fops);
1027         if (!ret)
1028                 list_add(&call->list, &ftrace_events);
1029
1030         return ret;
1031 }
1032
1033 /* Add an additional event_call dynamically */
1034 int trace_add_event_call(struct ftrace_event_call *call)
1035 {
1036         int ret;
1037         mutex_lock(&event_mutex);
1038         ret = __trace_add_event_call(call);
1039         mutex_unlock(&event_mutex);
1040         return ret;
1041 }
1042
1043 static void remove_subsystem_dir(const char *name)
1044 {
1045         struct event_subsystem *system;
1046
1047         if (strcmp(name, TRACE_SYSTEM) == 0)
1048                 return;
1049
1050         list_for_each_entry(system, &event_subsystems, list) {
1051                 if (strcmp(system->name, name) == 0) {
1052                         if (!--system->nr_events) {
1053                                 struct event_filter *filter = system->filter;
1054
1055                                 debugfs_remove_recursive(system->entry);
1056                                 list_del(&system->list);
1057                                 if (filter) {
1058                                         kfree(filter->filter_string);
1059                                         kfree(filter);
1060                                 }
1061                                 kfree(system->name);
1062                                 kfree(system);
1063                         }
1064                         break;
1065                 }
1066         }
1067 }
1068
1069 /*
1070  * Must be called under locking both of event_mutex and trace_event_mutex.
1071  */
1072 static void __trace_remove_event_call(struct ftrace_event_call *call)
1073 {
1074         ftrace_event_enable_disable(call, 0);
1075         if (call->event.funcs)
1076                 __unregister_ftrace_event(&call->event);
1077         debugfs_remove_recursive(call->dir);
1078         list_del(&call->list);
1079         trace_destroy_fields(call);
1080         destroy_preds(call);
1081         remove_subsystem_dir(call->class->system);
1082 }
1083
1084 /* Remove an event_call */
1085 void trace_remove_event_call(struct ftrace_event_call *call)
1086 {
1087         mutex_lock(&event_mutex);
1088         down_write(&trace_event_mutex);
1089         __trace_remove_event_call(call);
1090         up_write(&trace_event_mutex);
1091         mutex_unlock(&event_mutex);
1092 }
1093
1094 #define for_each_event(event, start, end)                       \
1095         for (event = start;                                     \
1096              (unsigned long)event < (unsigned long)end;         \
1097              event++)
1098
1099 #ifdef CONFIG_MODULES
1100
1101 static LIST_HEAD(ftrace_module_file_list);
1102
1103 /*
1104  * Modules must own their file_operations to keep up with
1105  * reference counting.
1106  */
1107 struct ftrace_module_file_ops {
1108         struct list_head                list;
1109         struct module                   *mod;
1110         struct file_operations          id;
1111         struct file_operations          enable;
1112         struct file_operations          format;
1113         struct file_operations          filter;
1114 };
1115
1116 static struct ftrace_module_file_ops *
1117 trace_create_file_ops(struct module *mod)
1118 {
1119         struct ftrace_module_file_ops *file_ops;
1120
1121         /*
1122          * This is a bit of a PITA. To allow for correct reference
1123          * counting, modules must "own" their file_operations.
1124          * To do this, we allocate the file operations that will be
1125          * used in the event directory.
1126          */
1127
1128         file_ops = kmalloc(sizeof(*file_ops), GFP_KERNEL);
1129         if (!file_ops)
1130                 return NULL;
1131
1132         file_ops->mod = mod;
1133
1134         file_ops->id = ftrace_event_id_fops;
1135         file_ops->id.owner = mod;
1136
1137         file_ops->enable = ftrace_enable_fops;
1138         file_ops->enable.owner = mod;
1139
1140         file_ops->filter = ftrace_event_filter_fops;
1141         file_ops->filter.owner = mod;
1142
1143         file_ops->format = ftrace_event_format_fops;
1144         file_ops->format.owner = mod;
1145
1146         list_add(&file_ops->list, &ftrace_module_file_list);
1147
1148         return file_ops;
1149 }
1150
1151 static void trace_module_add_events(struct module *mod)
1152 {
1153         struct ftrace_module_file_ops *file_ops = NULL;
1154         struct ftrace_event_call *call, *start, *end;
1155         struct dentry *d_events;
1156         int ret;
1157
1158         start = mod->trace_events;
1159         end = mod->trace_events + mod->num_trace_events;
1160
1161         if (start == end)
1162                 return;
1163
1164         d_events = event_trace_events_dir();
1165         if (!d_events)
1166                 return;
1167
1168         for_each_event(call, start, end) {
1169                 /* The linker may leave blanks */
1170                 if (!call->name)
1171                         continue;
1172                 if (call->class->raw_init) {
1173                         ret = call->class->raw_init(call);
1174                         if (ret < 0) {
1175                                 if (ret != -ENOSYS)
1176                                         pr_warning("Could not initialize trace "
1177                                         "point events/%s\n", call->name);
1178                                 continue;
1179                         }
1180                 }
1181                 /*
1182                  * This module has events, create file ops for this module
1183                  * if not already done.
1184                  */
1185                 if (!file_ops) {
1186                         file_ops = trace_create_file_ops(mod);
1187                         if (!file_ops)
1188                                 return;
1189                 }
1190                 call->mod = mod;
1191                 ret = event_create_dir(call, d_events,
1192                                        &file_ops->id, &file_ops->enable,
1193                                        &file_ops->filter, &file_ops->format);
1194                 if (!ret)
1195                         list_add(&call->list, &ftrace_events);
1196         }
1197 }
1198
1199 static void trace_module_remove_events(struct module *mod)
1200 {
1201         struct ftrace_module_file_ops *file_ops;
1202         struct ftrace_event_call *call, *p;
1203         bool found = false;
1204
1205         down_write(&trace_event_mutex);
1206         list_for_each_entry_safe(call, p, &ftrace_events, list) {
1207                 if (call->mod == mod) {
1208                         found = true;
1209                         __trace_remove_event_call(call);
1210                 }
1211         }
1212
1213         /* Now free the file_operations */
1214         list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1215                 if (file_ops->mod == mod)
1216                         break;
1217         }
1218         if (&file_ops->list != &ftrace_module_file_list) {
1219                 list_del(&file_ops->list);
1220                 kfree(file_ops);
1221         }
1222
1223         /*
1224          * It is safest to reset the ring buffer if the module being unloaded
1225          * registered any events.
1226          */
1227         if (found)
1228                 tracing_reset_current_online_cpus();
1229         up_write(&trace_event_mutex);
1230 }
1231
1232 static int trace_module_notify(struct notifier_block *self,
1233                                unsigned long val, void *data)
1234 {
1235         struct module *mod = data;
1236
1237         mutex_lock(&event_mutex);
1238         switch (val) {
1239         case MODULE_STATE_COMING:
1240                 trace_module_add_events(mod);
1241                 break;
1242         case MODULE_STATE_GOING:
1243                 trace_module_remove_events(mod);
1244                 break;
1245         }
1246         mutex_unlock(&event_mutex);
1247
1248         return 0;
1249 }
1250 #else
1251 static int trace_module_notify(struct notifier_block *self,
1252                                unsigned long val, void *data)
1253 {
1254         return 0;
1255 }
1256 #endif /* CONFIG_MODULES */
1257
1258 static struct notifier_block trace_module_nb = {
1259         .notifier_call = trace_module_notify,
1260         .priority = 0,
1261 };
1262
1263 extern struct ftrace_event_call __start_ftrace_events[];
1264 extern struct ftrace_event_call __stop_ftrace_events[];
1265
1266 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
1267
1268 static __init int setup_trace_event(char *str)
1269 {
1270         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
1271         ring_buffer_expanded = 1;
1272         tracing_selftest_disabled = 1;
1273
1274         return 1;
1275 }
1276 __setup("trace_event=", setup_trace_event);
1277
1278 static __init int event_trace_init(void)
1279 {
1280         struct ftrace_event_call *call;
1281         struct dentry *d_tracer;
1282         struct dentry *entry;
1283         struct dentry *d_events;
1284         int ret;
1285         char *buf = bootup_event_buf;
1286         char *token;
1287
1288         d_tracer = tracing_init_dentry();
1289         if (!d_tracer)
1290                 return 0;
1291
1292         entry = debugfs_create_file("available_events", 0444, d_tracer,
1293                                     (void *)&show_event_seq_ops,
1294                                     &ftrace_avail_fops);
1295         if (!entry)
1296                 pr_warning("Could not create debugfs "
1297                            "'available_events' entry\n");
1298
1299         entry = debugfs_create_file("set_event", 0644, d_tracer,
1300                                     (void *)&show_set_event_seq_ops,
1301                                     &ftrace_set_event_fops);
1302         if (!entry)
1303                 pr_warning("Could not create debugfs "
1304                            "'set_event' entry\n");
1305
1306         d_events = event_trace_events_dir();
1307         if (!d_events)
1308                 return 0;
1309
1310         /* ring buffer internal formats */
1311         trace_create_file("header_page", 0444, d_events,
1312                           ring_buffer_print_page_header,
1313                           &ftrace_show_header_fops);
1314
1315         trace_create_file("header_event", 0444, d_events,
1316                           ring_buffer_print_entry_header,
1317                           &ftrace_show_header_fops);
1318
1319         trace_create_file("enable", 0644, d_events,
1320                           NULL, &ftrace_system_enable_fops);
1321
1322         for_each_event(call, __start_ftrace_events, __stop_ftrace_events) {
1323                 /* The linker may leave blanks */
1324                 if (!call->name)
1325                         continue;
1326                 if (call->class->raw_init) {
1327                         ret = call->class->raw_init(call);
1328                         if (ret < 0) {
1329                                 if (ret != -ENOSYS)
1330                                         pr_warning("Could not initialize trace "
1331                                         "point events/%s\n", call->name);
1332                                 continue;
1333                         }
1334                 }
1335                 ret = event_create_dir(call, d_events, &ftrace_event_id_fops,
1336                                        &ftrace_enable_fops,
1337                                        &ftrace_event_filter_fops,
1338                                        &ftrace_event_format_fops);
1339                 if (!ret)
1340                         list_add(&call->list, &ftrace_events);
1341         }
1342
1343         while (true) {
1344                 token = strsep(&buf, ",");
1345
1346                 if (!token)
1347                         break;
1348                 if (!*token)
1349                         continue;
1350
1351                 ret = ftrace_set_clr_event(token, 1);
1352                 if (ret)
1353                         pr_warning("Failed to enable trace event: %s\n", token);
1354         }
1355
1356         ret = register_module_notifier(&trace_module_nb);
1357         if (ret)
1358                 pr_warning("Failed to register trace events module notifier\n");
1359
1360         return 0;
1361 }
1362 fs_initcall(event_trace_init);
1363
1364 #ifdef CONFIG_FTRACE_STARTUP_TEST
1365
1366 static DEFINE_SPINLOCK(test_spinlock);
1367 static DEFINE_SPINLOCK(test_spinlock_irq);
1368 static DEFINE_MUTEX(test_mutex);
1369
1370 static __init void test_work(struct work_struct *dummy)
1371 {
1372         spin_lock(&test_spinlock);
1373         spin_lock_irq(&test_spinlock_irq);
1374         udelay(1);
1375         spin_unlock_irq(&test_spinlock_irq);
1376         spin_unlock(&test_spinlock);
1377
1378         mutex_lock(&test_mutex);
1379         msleep(1);
1380         mutex_unlock(&test_mutex);
1381 }
1382
1383 static __init int event_test_thread(void *unused)
1384 {
1385         void *test_malloc;
1386
1387         test_malloc = kmalloc(1234, GFP_KERNEL);
1388         if (!test_malloc)
1389                 pr_info("failed to kmalloc\n");
1390
1391         schedule_on_each_cpu(test_work);
1392
1393         kfree(test_malloc);
1394
1395         set_current_state(TASK_INTERRUPTIBLE);
1396         while (!kthread_should_stop())
1397                 schedule();
1398
1399         return 0;
1400 }
1401
1402 /*
1403  * Do various things that may trigger events.
1404  */
1405 static __init void event_test_stuff(void)
1406 {
1407         struct task_struct *test_thread;
1408
1409         test_thread = kthread_run(event_test_thread, NULL, "test-events");
1410         msleep(1);
1411         kthread_stop(test_thread);
1412 }
1413
1414 /*
1415  * For every trace event defined, we will test each trace point separately,
1416  * and then by groups, and finally all trace points.
1417  */
1418 static __init void event_trace_self_tests(void)
1419 {
1420         struct ftrace_event_call *call;
1421         struct event_subsystem *system;
1422         int ret;
1423
1424         pr_info("Running tests on trace events:\n");
1425
1426         list_for_each_entry(call, &ftrace_events, list) {
1427
1428                 /* Only test those that have a probe */
1429                 if (!call->class || !call->class->probe)
1430                         continue;
1431
1432 /*
1433  * Testing syscall events here is pretty useless, but
1434  * we still do it if configured. But this is time consuming.
1435  * What we really need is a user thread to perform the
1436  * syscalls as we test.
1437  */
1438 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
1439                 if (call->class->system &&
1440                     strcmp(call->class->system, "syscalls") == 0)
1441                         continue;
1442 #endif
1443
1444                 pr_info("Testing event %s: ", call->name);
1445
1446                 /*
1447                  * If an event is already enabled, someone is using
1448                  * it and the self test should not be on.
1449                  */
1450                 if (call->flags & TRACE_EVENT_FL_ENABLED) {
1451                         pr_warning("Enabled event during self test!\n");
1452                         WARN_ON_ONCE(1);
1453                         continue;
1454                 }
1455
1456                 ftrace_event_enable_disable(call, 1);
1457                 event_test_stuff();
1458                 ftrace_event_enable_disable(call, 0);
1459
1460                 pr_cont("OK\n");
1461         }
1462
1463         /* Now test at the sub system level */
1464
1465         pr_info("Running tests on trace event systems:\n");
1466
1467         list_for_each_entry(system, &event_subsystems, list) {
1468
1469                 /* the ftrace system is special, skip it */
1470                 if (strcmp(system->name, "ftrace") == 0)
1471                         continue;
1472
1473                 pr_info("Testing event system %s: ", system->name);
1474
1475                 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 1);
1476                 if (WARN_ON_ONCE(ret)) {
1477                         pr_warning("error enabling system %s\n",
1478                                    system->name);
1479                         continue;
1480                 }
1481
1482                 event_test_stuff();
1483
1484                 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 0);
1485                 if (WARN_ON_ONCE(ret))
1486                         pr_warning("error disabling system %s\n",
1487                                    system->name);
1488
1489                 pr_cont("OK\n");
1490         }
1491
1492         /* Test with all events enabled */
1493
1494         pr_info("Running tests on all trace events:\n");
1495         pr_info("Testing all events: ");
1496
1497         ret = __ftrace_set_clr_event(NULL, NULL, NULL, 1);
1498         if (WARN_ON_ONCE(ret)) {
1499                 pr_warning("error enabling all events\n");
1500                 return;
1501         }
1502
1503         event_test_stuff();
1504
1505         /* reset sysname */
1506         ret = __ftrace_set_clr_event(NULL, NULL, NULL, 0);
1507         if (WARN_ON_ONCE(ret)) {
1508                 pr_warning("error disabling all events\n");
1509                 return;
1510         }
1511
1512         pr_cont("OK\n");
1513 }
1514
1515 #ifdef CONFIG_FUNCTION_TRACER
1516
1517 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
1518
1519 static void
1520 function_test_events_call(unsigned long ip, unsigned long parent_ip)
1521 {
1522         struct ring_buffer_event *event;
1523         struct ring_buffer *buffer;
1524         struct ftrace_entry *entry;
1525         unsigned long flags;
1526         long disabled;
1527         int resched;
1528         int cpu;
1529         int pc;
1530
1531         pc = preempt_count();
1532         resched = ftrace_preempt_disable();
1533         cpu = raw_smp_processor_id();
1534         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
1535
1536         if (disabled != 1)
1537                 goto out;
1538
1539         local_save_flags(flags);
1540
1541         event = trace_current_buffer_lock_reserve(&buffer,
1542                                                   TRACE_FN, sizeof(*entry),
1543                                                   flags, pc);
1544         if (!event)
1545                 goto out;
1546         entry   = ring_buffer_event_data(event);
1547         entry->ip                       = ip;
1548         entry->parent_ip                = parent_ip;
1549
1550         trace_nowake_buffer_unlock_commit(buffer, event, flags, pc);
1551
1552  out:
1553         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
1554         ftrace_preempt_enable(resched);
1555 }
1556
1557 static struct ftrace_ops trace_ops __initdata  =
1558 {
1559         .func = function_test_events_call,
1560 };
1561
1562 static __init void event_trace_self_test_with_function(void)
1563 {
1564         register_ftrace_function(&trace_ops);
1565         pr_info("Running tests again, along with the function tracer\n");
1566         event_trace_self_tests();
1567         unregister_ftrace_function(&trace_ops);
1568 }
1569 #else
1570 static __init void event_trace_self_test_with_function(void)
1571 {
1572 }
1573 #endif
1574
1575 static __init int event_trace_self_tests_init(void)
1576 {
1577         if (!tracing_selftest_disabled) {
1578                 event_trace_self_tests();
1579                 event_trace_self_test_with_function();
1580         }
1581
1582         return 0;
1583 }
1584
1585 late_initcall(event_trace_self_tests_init);
1586
1587 #endif