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         call->id = id;
 129
 130         return 0;
 131 }
 132 EXPORT_SYMBOL_GPL(trace_event_raw_init);
 133
 134 static int ftrace_event_enable_disable(struct ftrace_event_call *call,
 135                                         int enable)
 136 {
 137         int ret = 0;
 138
 139         switch (enable) {
 140         case 0:
 141                 if (call->enabled) {
 142                         call->enabled = 0;
 143                         tracing_stop_cmdline_record();
 144                         if (call->class->reg)
 145                                 call->class->reg(call, TRACE_REG_UNREGISTER);
 146                         else
 147                                 tracepoint_probe_unregister(call->name,
 148                                                             call->class->probe,
 149                                                             call);
 150                 }
 151                 break;
 152         case 1:
 153                 if (!call->enabled) {
 154                         tracing_start_cmdline_record();
 155                         if (call->class->reg)
 156                                 ret = call->class->reg(call, TRACE_REG_REGISTER);
 157                         else
 158                                 ret = tracepoint_probe_register(call->name,
 159                                                                 call->class->probe,
 160                                                                 call);
 161                         if (ret) {
 162                                 tracing_stop_cmdline_record();
 163                                 pr_info("event trace: Could not enable event "
 164                                         "%s\n", call->name);
 165                                 break;
 166                         }
 167                         call->enabled = 1;
 168                 }
 169                 break;
 170         }
 171
 172         return ret;
 173 }
 174
 175 static void ftrace_clear_events(void)
 176 {
 177         struct ftrace_event_call *call;
 178
 179         mutex_lock(&event_mutex);
 180         list_for_each_entry(call, &ftrace_events, list) {
 181                 ftrace_event_enable_disable(call, 0);
 182         }
 183         mutex_unlock(&event_mutex);
 184 }
 185
 186 /*
 187  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 188  */
 189 static int __ftrace_set_clr_event(const char *match, const char *sub,
 190                                   const char *event, int set)
 191 {
 192         struct ftrace_event_call *call;
 193         int ret = -EINVAL;
 194
 195         mutex_lock(&event_mutex);
 196         list_for_each_entry(call, &ftrace_events, list) {
 197
 198                 if (!call->name || !call->class ||
 199                     (!call->class->probe && !call->class->reg))
 200                         continue;
 201
 202                 if (match &&
 203                     strcmp(match, call->name) != 0 &&
 204                     strcmp(match, call->class->system) != 0)
 205                         continue;
 206
 207                 if (sub && strcmp(sub, call->class->system) != 0)
 208                         continue;
 209
 210                 if (event && strcmp(event, call->name) != 0)
 211                         continue;
 212
 213                 ftrace_event_enable_disable(call, set);
 214
 215                 ret = 0;
 216         }
 217         mutex_unlock(&event_mutex);
 218
 219         return ret;
 220 }
 221
 222 static int ftrace_set_clr_event(char *buf, int set)
 223 {
 224         char *event = NULL, *sub = NULL, *match;
 225
 226         /*
 227          * The buf format can be <subsystem>:<event-name>
 228          *  *:<event-name> means any event by that name.
 229          *  :<event-name> is the same.
 230          *
 231          *  <subsystem>:* means all events in that subsystem
 232          *  <subsystem>: means the same.
 233          *
 234          *  <name> (no ':') means all events in a subsystem with
 235          *  the name <name> or any event that matches <name>
 236          */
 237
 238         match = strsep(&buf, ":");
 239         if (buf) {
 240                 sub = match;
 241                 event = buf;
 242                 match = NULL;
 243
 244                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 245                         sub = NULL;
 246                 if (!strlen(event) || strcmp(event, "*") == 0)
 247                         event = NULL;
 248         }
 249
 250         return __ftrace_set_clr_event(match, sub, event, set);
 251 }
 252
 253 /**
 254  * trace_set_clr_event - enable or disable an event
 255  * @system: system name to match (NULL for any system)
 256  * @event: event name to match (NULL for all events, within system)
 257  * @set: 1 to enable, 0 to disable
 258  *
 259  * This is a way for other parts of the kernel to enable or disable
 260  * event recording.
 261  *
 262  * Returns 0 on success, -EINVAL if the parameters do not match any
 263  * registered events.
 264  */
 265 int trace_set_clr_event(const char *system, const char *event, int set)
 266 {
 267         return __ftrace_set_clr_event(NULL, system, event, set);
 268 }
 269
 270 /* 128 should be much more than enough */
 271 #define EVENT_BUF_SIZE          127
 272
 273 static ssize_t
 274 ftrace_event_write(struct file *file, const char __user *ubuf,
 275                    size_t cnt, loff_t *ppos)
 276 {
 277         struct trace_parser parser;
 278         ssize_t read, ret;
 279
 280         if (!cnt)
 281                 return 0;
 282
 283         ret = tracing_update_buffers();
 284         if (ret < 0)
 285                 return ret;
 286
 287         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 288                 return -ENOMEM;
 289
 290         read = trace_get_user(&parser, ubuf, cnt, ppos);
 291
 292         if (read >= 0 && trace_parser_loaded((&parser))) {
 293                 int set = 1;
 294
 295                 if (*parser.buffer == '!')
 296                         set = 0;
 297
 298                 parser.buffer[parser.idx] = 0;
 299
 300                 ret = ftrace_set_clr_event(parser.buffer + !set, set);
 301                 if (ret)
 302                         goto out_put;
 303         }
 304
 305         ret = read;
 306
 307  out_put:
 308         trace_parser_put(&parser);
 309
 310         return ret;
 311 }
 312
 313 static void *
 314 t_next(struct seq_file *m, void *v, loff_t *pos)
 315 {
 316         struct ftrace_event_call *call = v;
 317
 318         (*pos)++;
 319
 320         list_for_each_entry_continue(call, &ftrace_events, list) {
 321                 /*
 322                  * The ftrace subsystem is for showing formats only.
 323                  * They can not be enabled or disabled via the event files.
 324                  */
 325                 if (call->class && (call->class->probe || call->class->reg))
 326                         return call;
 327         }
 328
 329         return NULL;
 330 }
 331
 332 static void *t_start(struct seq_file *m, loff_t *pos)
 333 {
 334         struct ftrace_event_call *call;
 335         loff_t l;
 336
 337         mutex_lock(&event_mutex);
 338
 339         call = list_entry(&ftrace_events, struct ftrace_event_call, list);
 340         for (l = 0; l <= *pos; ) {
 341                 call = t_next(m, call, &l);
 342                 if (!call)
 343                         break;
 344         }
 345         return call;
 346 }
 347
 348 static void *
 349 s_next(struct seq_file *m, void *v, loff_t *pos)
 350 {
 351         struct ftrace_event_call *call = v;
 352
 353         (*pos)++;
 354
 355         list_for_each_entry_continue(call, &ftrace_events, list) {
 356                 if (call->enabled)
 357                         return call;
 358         }
 359
 360         return NULL;
 361 }
 362
 363 static void *s_start(struct seq_file *m, loff_t *pos)
 364 {
 365         struct ftrace_event_call *call;
 366         loff_t l;
 367
 368         mutex_lock(&event_mutex);
 369
 370         call = list_entry(&ftrace_events, struct ftrace_event_call, list);
 371         for (l = 0; l <= *pos; ) {
 372                 call = s_next(m, call, &l);
 373                 if (!call)
 374                         break;
 375         }
 376         return call;
 377 }
 378
 379 static int t_show(struct seq_file *m, void *v)
 380 {
 381         struct ftrace_event_call *call = v;
 382
 383         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
 384                 seq_printf(m, "%s:", call->class->system);
 385         seq_printf(m, "%s\n", call->name);
 386
 387         return 0;
 388 }
 389
 390 static void t_stop(struct seq_file *m, void *p)
 391 {
 392         mutex_unlock(&event_mutex);
 393 }
 394
 395 static int
 396 ftrace_event_seq_open(struct inode *inode, struct file *file)
 397 {
 398         const struct seq_operations *seq_ops;
 399
 400         if ((file->f_mode & FMODE_WRITE) &&
 401             (file->f_flags & O_TRUNC))
 402                 ftrace_clear_events();
 403
 404         seq_ops = inode->i_private;
 405         return seq_open(file, seq_ops);
 406 }
 407
 408 static ssize_t
 409 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 410                   loff_t *ppos)
 411 {
 412         struct ftrace_event_call *call = filp->private_data;
 413         char *buf;
 414
 415         if (call->enabled)
 416                 buf = "1\n";
 417         else
 418                 buf = "0\n";
 419
 420         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 421 }
 422
 423 static ssize_t
 424 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 425                    loff_t *ppos)
 426 {
 427         struct ftrace_event_call *call = filp->private_data;
 428         char buf[64];
 429         unsigned long val;
 430         int ret;
 431
 432         if (cnt >= sizeof(buf))
 433                 return -EINVAL;
 434
 435         if (copy_from_user(&buf, ubuf, cnt))
 436                 return -EFAULT;
 437
 438         buf[cnt] = 0;
 439
 440         ret = strict_strtoul(buf, 10, &val);
 441         if (ret < 0)
 442                 return ret;
 443
 444         ret = tracing_update_buffers();
 445         if (ret < 0)
 446                 return ret;
 447
 448         switch (val) {
 449         case 0:
 450         case 1:
 451                 mutex_lock(&event_mutex);
 452                 ret = ftrace_event_enable_disable(call, val);
 453                 mutex_unlock(&event_mutex);
 454                 break;
 455
 456         default:
 457                 return -EINVAL;
 458         }
 459
 460         *ppos += cnt;
 461
 462         return ret ? ret : cnt;
 463 }
 464
 465 static ssize_t
 466 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 467                    loff_t *ppos)
 468 {
 469         const char set_to_char[4] = { '?', '0', '1', 'X' };
 470         const char *system = filp->private_data;
 471         struct ftrace_event_call *call;
 472         char buf[2];
 473         int set = 0;
 474         int ret;
 475
 476         mutex_lock(&event_mutex);
 477         list_for_each_entry(call, &ftrace_events, list) {
 478                 if (!call->name || !call->class ||
 479                     (!call->class->probe && !call->class->reg))
 480                         continue;
 481
 482                 if (system && strcmp(call->class->system, system) != 0)
 483                         continue;
 484
 485                 /*
 486                  * We need to find out if all the events are set
 487                  * or if all events or cleared, or if we have
 488                  * a mixture.
 489                  */
 490                 set |= (1 << !!call->enabled);
 491
 492                 /*
 493                  * If we have a mixture, no need to look further.
 494                  */
 495                 if (set == 3)
 496                         break;
 497         }
 498         mutex_unlock(&event_mutex);
 499
 500         buf[0] = set_to_char[set];
 501         buf[1] = '\n';
 502
 503         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 504
 505         return ret;
 506 }
 507
 508 static ssize_t
 509 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 510                     loff_t *ppos)
 511 {
 512         const char *system = filp->private_data;
 513         unsigned long val;
 514         char buf[64];
 515         ssize_t ret;
 516
 517         if (cnt >= sizeof(buf))
 518                 return -EINVAL;
 519
 520         if (copy_from_user(&buf, ubuf, cnt))
 521                 return -EFAULT;
 522
 523         buf[cnt] = 0;
 524
 525         ret = strict_strtoul(buf, 10, &val);
 526         if (ret < 0)
 527                 return ret;
 528
 529         ret = tracing_update_buffers();
 530         if (ret < 0)
 531                 return ret;
 532
 533         if (val != 0 && val != 1)
 534                 return -EINVAL;
 535
 536         ret = __ftrace_set_clr_event(NULL, system, NULL, val);
 537         if (ret)
 538                 goto out;
 539
 540         ret = cnt;
 541
 542 out:
 543         *ppos += cnt;
 544
 545         return ret;
 546 }
 547
 548 static ssize_t
 549 event_format_read(struct file *filp, char __user *ubuf, size_t cnt,
 550                   loff_t *ppos)
 551 {
 552         struct ftrace_event_call *call = filp->private_data;
 553         struct ftrace_event_field *field;
 554         struct list_head *head;
 555         struct trace_seq *s;
 556         int common_field_count = 5;
 557         char *buf;
 558         int r = 0;
 559
 560         if (*ppos)
 561                 return 0;
 562
 563         s = kmalloc(sizeof(*s), GFP_KERNEL);
 564         if (!s)
 565                 return -ENOMEM;
 566
 567         trace_seq_init(s);
 568
 569         trace_seq_printf(s, "name: %s\n", call->name);
 570         trace_seq_printf(s, "ID: %d\n", call->id);
 571         trace_seq_printf(s, "format:\n");
 572
 573         head = trace_get_fields(call);
 574         list_for_each_entry_reverse(field, head, link) {
 575                 /*
 576                  * Smartly shows the array type(except dynamic array).
 577                  * Normal:
 578                  *      field:TYPE VAR
 579                  * If TYPE := TYPE[LEN], it is shown:
 580                  *      field:TYPE VAR[LEN]
 581                  */
 582                 const char *array_descriptor = strchr(field->type, '[');
 583
 584                 if (!strncmp(field->type, "__data_loc", 10))
 585                         array_descriptor = NULL;
 586
 587                 if (!array_descriptor) {
 588                         r = trace_seq_printf(s, "\tfield:%s %s;\toffset:%u;"
 589                                         "\tsize:%u;\tsigned:%d;\n",
 590                                         field->type, field->name, field->offset,
 591                                         field->size, !!field->is_signed);
 592                 } else {
 593                         r = trace_seq_printf(s, "\tfield:%.*s %s%s;\toffset:%u;"
 594                                         "\tsize:%u;\tsigned:%d;\n",
 595                                         (int)(array_descriptor - field->type),
 596                                         field->type, field->name,
 597                                         array_descriptor, field->offset,
 598                                         field->size, !!field->is_signed);
 599                 }
 600
 601                 if (--common_field_count == 0)
 602                         r = trace_seq_printf(s, "\n");
 603
 604                 if (!r)
 605                         break;
 606         }
 607
 608         if (r)
 609                 r = trace_seq_printf(s, "\nprint fmt: %s\n",
 610                                 call->print_fmt);
 611
 612         if (!r) {
 613                 /*
 614                  * ug!  The format output is bigger than a PAGE!!
 615                  */
 616                 buf = "FORMAT TOO BIG\n";
 617                 r = simple_read_from_buffer(ubuf, cnt, ppos,
 618                                               buf, strlen(buf));
 619                 goto out;
 620         }
 621
 622         r = simple_read_from_buffer(ubuf, cnt, ppos,
 623                                     s->buffer, s->len);
 624  out:
 625         kfree(s);
 626         return r;
 627 }
 628
 629 static ssize_t
 630 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 631 {
 632         struct ftrace_event_call *call = filp->private_data;
 633         struct trace_seq *s;
 634         int r;
 635
 636         if (*ppos)
 637                 return 0;
 638
 639         s = kmalloc(sizeof(*s), GFP_KERNEL);
 640         if (!s)
 641                 return -ENOMEM;
 642
 643         trace_seq_init(s);
 644         trace_seq_printf(s, "%d\n", call->id);
 645
 646         r = simple_read_from_buffer(ubuf, cnt, ppos,
 647                                     s->buffer, s->len);
 648         kfree(s);
 649         return r;
 650 }
 651
 652 static ssize_t
 653 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 654                   loff_t *ppos)
 655 {
 656         struct ftrace_event_call *call = filp->private_data;
 657         struct trace_seq *s;
 658         int r;
 659
 660         if (*ppos)
 661                 return 0;
 662
 663         s = kmalloc(sizeof(*s), GFP_KERNEL);
 664         if (!s)
 665                 return -ENOMEM;
 666
 667         trace_seq_init(s);
 668
 669         print_event_filter(call, s);
 670         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 671
 672         kfree(s);
 673
 674         return r;
 675 }
 676
 677 static ssize_t
 678 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
 679                    loff_t *ppos)
 680 {
 681         struct ftrace_event_call *call = filp->private_data;
 682         char *buf;
 683         int err;
 684
 685         if (cnt >= PAGE_SIZE)
 686                 return -EINVAL;
 687
 688         buf = (char *)__get_free_page(GFP_TEMPORARY);
 689         if (!buf)
 690                 return -ENOMEM;
 691
 692         if (copy_from_user(buf, ubuf, cnt)) {
 693                 free_page((unsigned long) buf);
 694                 return -EFAULT;
 695         }
 696         buf[cnt] = '\0';
 697
 698         err = apply_event_filter(call, buf);
 699         free_page((unsigned long) buf);
 700         if (err < 0)
 701                 return err;
 702
 703         *ppos += cnt;
 704
 705         return cnt;
 706 }
 707
 708 static ssize_t
 709 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 710                       loff_t *ppos)
 711 {
 712         struct event_subsystem *system = filp->private_data;
 713         struct trace_seq *s;
 714         int r;
 715
 716         if (*ppos)
 717                 return 0;
 718
 719         s = kmalloc(sizeof(*s), GFP_KERNEL);
 720         if (!s)
 721                 return -ENOMEM;
 722
 723         trace_seq_init(s);
 724
 725         print_subsystem_event_filter(system, s);
 726         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 727
 728         kfree(s);
 729
 730         return r;
 731 }
 732
 733 static ssize_t
 734 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
 735                        loff_t *ppos)
 736 {
 737         struct event_subsystem *system = filp->private_data;
 738         char *buf;
 739         int err;
 740
 741         if (cnt >= PAGE_SIZE)
 742                 return -EINVAL;
 743
 744         buf = (char *)__get_free_page(GFP_TEMPORARY);
 745         if (!buf)
 746                 return -ENOMEM;
 747
 748         if (copy_from_user(buf, ubuf, cnt)) {
 749                 free_page((unsigned long) buf);
 750                 return -EFAULT;
 751         }
 752         buf[cnt] = '\0';
 753
 754         err = apply_subsystem_event_filter(system, buf);
 755         free_page((unsigned long) buf);
 756         if (err < 0)
 757                 return err;
 758
 759         *ppos += cnt;
 760
 761         return cnt;
 762 }
 763
 764 static ssize_t
 765 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 766 {
 767         int (*func)(struct trace_seq *s) = filp->private_data;
 768         struct trace_seq *s;
 769         int r;
 770
 771         if (*ppos)
 772                 return 0;
 773
 774         s = kmalloc(sizeof(*s), GFP_KERNEL);
 775         if (!s)
 776                 return -ENOMEM;
 777
 778         trace_seq_init(s);
 779
 780         func(s);
 781         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 782
 783         kfree(s);
 784
 785         return r;
 786 }
 787
 788 static const struct seq_operations show_event_seq_ops = {
 789         .start = t_start,
 790         .next = t_next,
 791         .show = t_show,
 792         .stop = t_stop,
 793 };
 794
 795 static const struct seq_operations show_set_event_seq_ops = {
 796         .start = s_start,
 797         .next = s_next,
 798         .show = t_show,
 799         .stop = t_stop,
 800 };
 801
 802 static const struct file_operations ftrace_avail_fops = {
 803         .open = ftrace_event_seq_open,
 804         .read = seq_read,
 805         .llseek = seq_lseek,
 806         .release = seq_release,
 807 };
 808
 809 static const struct file_operations ftrace_set_event_fops = {
 810         .open = ftrace_event_seq_open,
 811         .read = seq_read,
 812         .write = ftrace_event_write,
 813         .llseek = seq_lseek,
 814         .release = seq_release,
 815 };
 816
 817 static const struct file_operations ftrace_enable_fops = {
 818         .open = tracing_open_generic,
 819         .read = event_enable_read,
 820         .write = event_enable_write,
 821 };
 822
 823 static const struct file_operations ftrace_event_format_fops = {
 824         .open = tracing_open_generic,
 825         .read = event_format_read,
 826 };
 827
 828 static const struct file_operations ftrace_event_id_fops = {
 829         .open = tracing_open_generic,
 830         .read = event_id_read,
 831 };
 832
 833 static const struct file_operations ftrace_event_filter_fops = {
 834         .open = tracing_open_generic,
 835         .read = event_filter_read,
 836         .write = event_filter_write,
 837 };
 838
 839 static const struct file_operations ftrace_subsystem_filter_fops = {
 840         .open = tracing_open_generic,
 841         .read = subsystem_filter_read,
 842         .write = subsystem_filter_write,
 843 };
 844
 845 static const struct file_operations ftrace_system_enable_fops = {
 846         .open = tracing_open_generic,
 847         .read = system_enable_read,
 848         .write = system_enable_write,
 849 };
 850
 851 static const struct file_operations ftrace_show_header_fops = {
 852         .open = tracing_open_generic,
 853         .read = show_header,
 854 };
 855
 856 static struct dentry *event_trace_events_dir(void)
 857 {
 858         static struct dentry *d_tracer;
 859         static struct dentry *d_events;
 860
 861         if (d_events)
 862                 return d_events;
 863
 864         d_tracer = tracing_init_dentry();
 865         if (!d_tracer)
 866                 return NULL;
 867
 868         d_events = debugfs_create_dir("events", d_tracer);
 869         if (!d_events)
 870                 pr_warning("Could not create debugfs "
 871                            "'events' directory\n");
 872
 873         return d_events;
 874 }
 875
 876 static LIST_HEAD(event_subsystems);
 877
 878 static struct dentry *
 879 event_subsystem_dir(const char *name, struct dentry *d_events)
 880 {
 881         struct event_subsystem *system;
 882         struct dentry *entry;
 883
 884         /* First see if we did not already create this dir */
 885         list_for_each_entry(system, &event_subsystems, list) {
 886                 if (strcmp(system->name, name) == 0) {
 887                         system->nr_events++;
 888                         return system->entry;
 889                 }
 890         }
 891
 892         /* need to create new entry */
 893         system = kmalloc(sizeof(*system), GFP_KERNEL);
 894         if (!system) {
 895                 pr_warning("No memory to create event subsystem %s\n",
 896                            name);
 897                 return d_events;
 898         }
 899
 900         system->entry = debugfs_create_dir(name, d_events);
 901         if (!system->entry) {
 902                 pr_warning("Could not create event subsystem %s\n",
 903                            name);
 904                 kfree(system);
 905                 return d_events;
 906         }
 907
 908         system->nr_events = 1;
 909         system->name = kstrdup(name, GFP_KERNEL);
 910         if (!system->name) {
 911                 debugfs_remove(system->entry);
 912                 kfree(system);
 913                 return d_events;
 914         }
 915
 916         list_add(&system->list, &event_subsystems);
 917
 918         system->filter = NULL;
 919
 920         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
 921         if (!system->filter) {
 922                 pr_warning("Could not allocate filter for subsystem "
 923                            "'%s'\n", name);
 924                 return system->entry;
 925         }
 926
 927         entry = debugfs_create_file("filter", 0644, system->entry, system,
 928                                     &ftrace_subsystem_filter_fops);
 929         if (!entry) {
 930                 kfree(system->filter);
 931                 system->filter = NULL;
 932                 pr_warning("Could not create debugfs "
 933                            "'%s/filter' entry\n", name);
 934         }
 935
 936         trace_create_file("enable", 0644, system->entry,
 937                           (void *)system->name,
 938                           &ftrace_system_enable_fops);
 939
 940         return system->entry;
 941 }
 942
 943 static int
 944 event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
 945                  const struct file_operations *id,
 946                  const struct file_operations *enable,
 947                  const struct file_operations *filter,
 948                  const struct file_operations *format)
 949 {
 950         struct list_head *head;
 951         int ret;
 952
 953         /*
 954          * If the trace point header did not define TRACE_SYSTEM
 955          * then the system would be called "TRACE_SYSTEM".
 956          */
 957         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
 958                 d_events = event_subsystem_dir(call->class->system, d_events);
 959
 960         call->dir = debugfs_create_dir(call->name, d_events);
 961         if (!call->dir) {
 962                 pr_warning("Could not create debugfs "
 963                            "'%s' directory\n", call->name);
 964                 return -1;
 965         }
 966
 967         if (call->class->probe || call->class->reg)
 968                 trace_create_file("enable", 0644, call->dir, call,
 969                                   enable);
 970
 971 #ifdef CONFIG_PERF_EVENTS
 972         if (call->id && (call->class->perf_probe || call->class->reg))
 973                 trace_create_file("id", 0444, call->dir, call,
 974                                   id);
 975 #endif
 976
 977         if (call->class->define_fields) {
 978                 /*
 979                  * Other events may have the same class. Only update
 980                  * the fields if they are not already defined.
 981                  */
 982                 head = trace_get_fields(call);
 983                 if (list_empty(head)) {
 984                         ret = trace_define_common_fields(call);
 985                         if (!ret)
 986                                 ret = call->class->define_fields(call);
 987                         if (ret < 0) {
 988                                 pr_warning("Could not initialize trace point"
 989                                            " events/%s\n", call->name);
 990                                 return ret;
 991                         }
 992                 }
 993                 trace_create_file("filter", 0644, call->dir, call,
 994                                   filter);
 995         }
 996
 997         trace_create_file("format", 0444, call->dir, call,
 998                           format);
 999
1000         return 0;
1001 }
1002
1003 static int __trace_add_event_call(struct ftrace_event_call *call)
1004 {
1005         struct dentry *d_events;
1006         int ret;
1007
1008         if (!call->name)
1009                 return -EINVAL;
1010
1011         if (call->class->raw_init) {
1012                 ret = call->class->raw_init(call);
1013                 if (ret < 0) {
1014                         if (ret != -ENOSYS)
1015                                 pr_warning("Could not initialize trace "
1016                                 "events/%s\n", call->name);
1017                         return ret;
1018                 }
1019         }
1020
1021         d_events = event_trace_events_dir();
1022         if (!d_events)
1023                 return -ENOENT;
1024
1025         ret = event_create_dir(call, d_events, &ftrace_event_id_fops,
1026                                 &ftrace_enable_fops, &ftrace_event_filter_fops,
1027                                 &ftrace_event_format_fops);
1028         if (!ret)
1029                 list_add(&call->list, &ftrace_events);
1030
1031         return ret;
1032 }
1033
1034 /* Add an additional event_call dynamically */
1035 int trace_add_event_call(struct ftrace_event_call *call)
1036 {
1037         int ret;
1038         mutex_lock(&event_mutex);
1039         ret = __trace_add_event_call(call);
1040         mutex_unlock(&event_mutex);
1041         return ret;
1042 }
1043
1044 static void remove_subsystem_dir(const char *name)
1045 {
1046         struct event_subsystem *system;
1047
1048         if (strcmp(name, TRACE_SYSTEM) == 0)
1049                 return;
1050
1051         list_for_each_entry(system, &event_subsystems, list) {
1052                 if (strcmp(system->name, name) == 0) {
1053                         if (!--system->nr_events) {
1054                                 struct event_filter *filter = system->filter;
1055
1056                                 debugfs_remove_recursive(system->entry);
1057                                 list_del(&system->list);
1058                                 if (filter) {
1059                                         kfree(filter->filter_string);
1060                                         kfree(filter);
1061                                 }
1062                                 kfree(system->name);
1063                                 kfree(system);
1064                         }
1065                         break;
1066                 }
1067         }
1068 }
1069
1070 /*
1071  * Must be called under locking both of event_mutex and trace_event_mutex.
1072  */
1073 static void __trace_remove_event_call(struct ftrace_event_call *call)
1074 {
1075         ftrace_event_enable_disable(call, 0);
1076         if (call->event.funcs)
1077                 __unregister_ftrace_event(&call->event);
1078         debugfs_remove_recursive(call->dir);
1079         list_del(&call->list);
1080         trace_destroy_fields(call);
1081         destroy_preds(call);
1082         remove_subsystem_dir(call->class->system);
1083 }
1084
1085 /* Remove an event_call */
1086 void trace_remove_event_call(struct ftrace_event_call *call)
1087 {
1088         mutex_lock(&event_mutex);
1089         down_write(&trace_event_mutex);
1090         __trace_remove_event_call(call);
1091         up_write(&trace_event_mutex);
1092         mutex_unlock(&event_mutex);
1093 }
1094
1095 #define for_each_event(event, start, end)                       \
1096         for (event = start;                                     \
1097              (unsigned long)event < (unsigned long)end;         \
1098              event++)
1099
1100 #ifdef CONFIG_MODULES
1101
1102 static LIST_HEAD(ftrace_module_file_list);
1103
1104 /*
1105  * Modules must own their file_operations to keep up with
1106  * reference counting.
1107  */
1108 struct ftrace_module_file_ops {
1109         struct list_head                list;
1110         struct module                   *mod;
1111         struct file_operations          id;
1112         struct file_operations          enable;
1113         struct file_operations          format;
1114         struct file_operations          filter;
1115 };
1116
1117 static struct ftrace_module_file_ops *
1118 trace_create_file_ops(struct module *mod)
1119 {
1120         struct ftrace_module_file_ops *file_ops;
1121
1122         /*
1123          * This is a bit of a PITA. To allow for correct reference
1124          * counting, modules must "own" their file_operations.
1125          * To do this, we allocate the file operations that will be
1126          * used in the event directory.
1127          */
1128
1129         file_ops = kmalloc(sizeof(*file_ops), GFP_KERNEL);
1130         if (!file_ops)
1131                 return NULL;
1132
1133         file_ops->mod = mod;
1134
1135         file_ops->id = ftrace_event_id_fops;
1136         file_ops->id.owner = mod;
1137
1138         file_ops->enable = ftrace_enable_fops;
1139         file_ops->enable.owner = mod;
1140
1141         file_ops->filter = ftrace_event_filter_fops;
1142         file_ops->filter.owner = mod;
1143
1144         file_ops->format = ftrace_event_format_fops;
1145         file_ops->format.owner = mod;
1146
1147         list_add(&file_ops->list, &ftrace_module_file_list);
1148
1149         return file_ops;
1150 }
1151
1152 static void trace_module_add_events(struct module *mod)
1153 {
1154         struct ftrace_module_file_ops *file_ops = NULL;
1155         struct ftrace_event_call *call, *start, *end;
1156         struct dentry *d_events;
1157         int ret;
1158
1159         start = mod->trace_events;
1160         end = mod->trace_events + mod->num_trace_events;
1161
1162         if (start == end)
1163                 return;
1164
1165         d_events = event_trace_events_dir();
1166         if (!d_events)
1167                 return;
1168
1169         for_each_event(call, start, end) {
1170                 /* The linker may leave blanks */
1171                 if (!call->name)
1172                         continue;
1173                 if (call->class->raw_init) {
1174                         ret = call->class->raw_init(call);
1175                         if (ret < 0) {
1176                                 if (ret != -ENOSYS)
1177                                         pr_warning("Could not initialize trace "
1178                                         "point events/%s\n", call->name);
1179                                 continue;
1180                         }
1181                 }
1182                 /*
1183                  * This module has events, create file ops for this module
1184                  * if not already done.
1185                  */
1186                 if (!file_ops) {
1187                         file_ops = trace_create_file_ops(mod);
1188                         if (!file_ops)
1189                                 return;
1190                 }
1191                 call->mod = mod;
1192                 ret = event_create_dir(call, d_events,
1193                                        &file_ops->id, &file_ops->enable,
1194                                        &file_ops->filter, &file_ops->format);
1195                 if (!ret)
1196                         list_add(&call->list, &ftrace_events);
1197         }
1198 }
1199
1200 static void trace_module_remove_events(struct module *mod)
1201 {
1202         struct ftrace_module_file_ops *file_ops;
1203         struct ftrace_event_call *call, *p;
1204         bool found = false;
1205
1206         down_write(&trace_event_mutex);
1207         list_for_each_entry_safe(call, p, &ftrace_events, list) {
1208                 if (call->mod == mod) {
1209                         found = true;
1210                         __trace_remove_event_call(call);
1211                 }
1212         }
1213
1214         /* Now free the file_operations */
1215         list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1216                 if (file_ops->mod == mod)
1217                         break;
1218         }
1219         if (&file_ops->list != &ftrace_module_file_list) {
1220                 list_del(&file_ops->list);
1221                 kfree(file_ops);
1222         }
1223
1224         /*
1225          * It is safest to reset the ring buffer if the module being unloaded
1226          * registered any events.
1227          */
1228         if (found)
1229                 tracing_reset_current_online_cpus();
1230         up_write(&trace_event_mutex);
1231 }
1232
1233 static int trace_module_notify(struct notifier_block *self,
1234                                unsigned long val, void *data)
1235 {
1236         struct module *mod = data;
1237
1238         mutex_lock(&event_mutex);
1239         switch (val) {
1240         case MODULE_STATE_COMING:
1241                 trace_module_add_events(mod);
1242                 break;
1243         case MODULE_STATE_GOING:
1244                 trace_module_remove_events(mod);
1245                 break;
1246         }
1247         mutex_unlock(&event_mutex);
1248
1249         return 0;
1250 }
1251 #else
1252 static int trace_module_notify(struct notifier_block *self,
1253                                unsigned long val, void *data)
1254 {
1255         return 0;
1256 }
1257 #endif /* CONFIG_MODULES */
1258
1259 static struct notifier_block trace_module_nb = {
1260         .notifier_call = trace_module_notify,
1261         .priority = 0,
1262 };
1263
1264 extern struct ftrace_event_call __start_ftrace_events[];
1265 extern struct ftrace_event_call __stop_ftrace_events[];
1266
1267 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
1268
1269 static __init int setup_trace_event(char *str)
1270 {
1271         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
1272         ring_buffer_expanded = 1;
1273         tracing_selftest_disabled = 1;
1274
1275         return 1;
1276 }
1277 __setup("trace_event=", setup_trace_event);
1278
1279 static __init int event_trace_init(void)
1280 {
1281         struct ftrace_event_call *call;
1282         struct dentry *d_tracer;
1283         struct dentry *entry;
1284         struct dentry *d_events;
1285         int ret;
1286         char *buf = bootup_event_buf;
1287         char *token;
1288
1289         d_tracer = tracing_init_dentry();
1290         if (!d_tracer)
1291                 return 0;
1292
1293         entry = debugfs_create_file("available_events", 0444, d_tracer,
1294                                     (void *)&show_event_seq_ops,
1295                                     &ftrace_avail_fops);
1296         if (!entry)
1297                 pr_warning("Could not create debugfs "
1298                            "'available_events' entry\n");
1299
1300         entry = debugfs_create_file("set_event", 0644, d_tracer,
1301                                     (void *)&show_set_event_seq_ops,
1302                                     &ftrace_set_event_fops);
1303         if (!entry)
1304                 pr_warning("Could not create debugfs "
1305                            "'set_event' entry\n");
1306
1307         d_events = event_trace_events_dir();
1308         if (!d_events)
1309                 return 0;
1310
1311         /* ring buffer internal formats */
1312         trace_create_file("header_page", 0444, d_events,
1313                           ring_buffer_print_page_header,
1314                           &ftrace_show_header_fops);
1315
1316         trace_create_file("header_event", 0444, d_events,
1317                           ring_buffer_print_entry_header,
1318                           &ftrace_show_header_fops);
1319
1320         trace_create_file("enable", 0644, d_events,
1321                           NULL, &ftrace_system_enable_fops);
1322
1323         for_each_event(call, __start_ftrace_events, __stop_ftrace_events) {
1324                 /* The linker may leave blanks */
1325                 if (!call->name)
1326                         continue;
1327                 if (call->class->raw_init) {
1328                         ret = call->class->raw_init(call);
1329                         if (ret < 0) {
1330                                 if (ret != -ENOSYS)
1331                                         pr_warning("Could not initialize trace "
1332                                         "point events/%s\n", call->name);
1333                                 continue;
1334                         }
1335                 }
1336                 ret = event_create_dir(call, d_events, &ftrace_event_id_fops,
1337                                        &ftrace_enable_fops,
1338                                        &ftrace_event_filter_fops,
1339                                        &ftrace_event_format_fops);
1340                 if (!ret)
1341                         list_add(&call->list, &ftrace_events);
1342         }
1343
1344         while (true) {
1345                 token = strsep(&buf, ",");
1346
1347                 if (!token)
1348                         break;
1349                 if (!*token)
1350                         continue;
1351
1352                 ret = ftrace_set_clr_event(token, 1);
1353                 if (ret)
1354                         pr_warning("Failed to enable trace event: %s\n", token);
1355         }
1356
1357         ret = register_module_notifier(&trace_module_nb);
1358         if (ret)
1359                 pr_warning("Failed to register trace events module notifier\n");
1360
1361         return 0;
1362 }
1363 fs_initcall(event_trace_init);
1364
1365 #ifdef CONFIG_FTRACE_STARTUP_TEST
1366
1367 static DEFINE_SPINLOCK(test_spinlock);
1368 static DEFINE_SPINLOCK(test_spinlock_irq);
1369 static DEFINE_MUTEX(test_mutex);
1370
1371 static __init void test_work(struct work_struct *dummy)
1372 {
1373         spin_lock(&test_spinlock);
1374         spin_lock_irq(&test_spinlock_irq);
1375         udelay(1);
1376         spin_unlock_irq(&test_spinlock_irq);
1377         spin_unlock(&test_spinlock);
1378
1379         mutex_lock(&test_mutex);
1380         msleep(1);
1381         mutex_unlock(&test_mutex);
1382 }
1383
1384 static __init int event_test_thread(void *unused)
1385 {
1386         void *test_malloc;
1387
1388         test_malloc = kmalloc(1234, GFP_KERNEL);
1389         if (!test_malloc)
1390                 pr_info("failed to kmalloc\n");
1391
1392         schedule_on_each_cpu(test_work);
1393
1394         kfree(test_malloc);
1395
1396         set_current_state(TASK_INTERRUPTIBLE);
1397         while (!kthread_should_stop())
1398                 schedule();
1399
1400         return 0;
1401 }
1402
1403 /*
1404  * Do various things that may trigger events.
1405  */
1406 static __init void event_test_stuff(void)
1407 {
1408         struct task_struct *test_thread;
1409
1410         test_thread = kthread_run(event_test_thread, NULL, "test-events");
1411         msleep(1);
1412         kthread_stop(test_thread);
1413 }
1414
1415 /*
1416  * For every trace event defined, we will test each trace point separately,
1417  * and then by groups, and finally all trace points.
1418  */
1419 static __init void event_trace_self_tests(void)
1420 {
1421         struct ftrace_event_call *call;
1422         struct event_subsystem *system;
1423         int ret;
1424
1425         pr_info("Running tests on trace events:\n");
1426
1427         list_for_each_entry(call, &ftrace_events, list) {
1428
1429                 /* Only test those that have a probe */
1430                 if (!call->class || !call->class->probe)
1431                         continue;
1432
1433 /*
1434  * Testing syscall events here is pretty useless, but
1435  * we still do it if configured. But this is time consuming.
1436  * What we really need is a user thread to perform the
1437  * syscalls as we test.
1438  */
1439 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
1440                 if (call->class->system &&
1441                     strcmp(call->class->system, "syscalls") == 0)
1442                         continue;
1443 #endif
1444
1445                 pr_info("Testing event %s: ", call->name);
1446
1447                 /*
1448                  * If an event is already enabled, someone is using
1449                  * it and the self test should not be on.
1450                  */
1451                 if (call->enabled) {
1452                         pr_warning("Enabled event during self test!\n");
1453                         WARN_ON_ONCE(1);
1454                         continue;
1455                 }
1456
1457                 ftrace_event_enable_disable(call, 1);
1458                 event_test_stuff();
1459                 ftrace_event_enable_disable(call, 0);
1460
1461                 pr_cont("OK\n");
1462         }
1463
1464         /* Now test at the sub system level */
1465
1466         pr_info("Running tests on trace event systems:\n");
1467
1468         list_for_each_entry(system, &event_subsystems, list) {
1469
1470                 /* the ftrace system is special, skip it */
1471                 if (strcmp(system->name, "ftrace") == 0)
1472                         continue;
1473
1474                 pr_info("Testing event system %s: ", system->name);
1475
1476                 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 1);
1477                 if (WARN_ON_ONCE(ret)) {
1478                         pr_warning("error enabling system %s\n",
1479                                    system->name);
1480                         continue;
1481                 }
1482
1483                 event_test_stuff();
1484
1485                 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 0);
1486                 if (WARN_ON_ONCE(ret))
1487                         pr_warning("error disabling system %s\n",
1488                                    system->name);
1489
1490                 pr_cont("OK\n");
1491         }
1492
1493         /* Test with all events enabled */
1494
1495         pr_info("Running tests on all trace events:\n");
1496         pr_info("Testing all events: ");
1497
1498         ret = __ftrace_set_clr_event(NULL, NULL, NULL, 1);
1499         if (WARN_ON_ONCE(ret)) {
1500                 pr_warning("error enabling all events\n");
1501                 return;
1502         }
1503
1504         event_test_stuff();
1505
1506         /* reset sysname */
1507         ret = __ftrace_set_clr_event(NULL, NULL, NULL, 0);
1508         if (WARN_ON_ONCE(ret)) {
1509                 pr_warning("error disabling all events\n");
1510                 return;
1511         }
1512
1513         pr_cont("OK\n");
1514 }
1515
1516 #ifdef CONFIG_FUNCTION_TRACER
1517
1518 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
1519
1520 static void
1521 function_test_events_call(unsigned long ip, unsigned long parent_ip)
1522 {
1523         struct ring_buffer_event *event;
1524         struct ring_buffer *buffer;
1525         struct ftrace_entry *entry;
1526         unsigned long flags;
1527         long disabled;
1528         int resched;
1529         int cpu;
1530         int pc;
1531
1532         pc = preempt_count();
1533         resched = ftrace_preempt_disable();
1534         cpu = raw_smp_processor_id();
1535         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
1536
1537         if (disabled != 1)
1538                 goto out;
1539
1540         local_save_flags(flags);
1541
1542         event = trace_current_buffer_lock_reserve(&buffer,
1543                                                   TRACE_FN, sizeof(*entry),
1544                                                   flags, pc);
1545         if (!event)
1546                 goto out;
1547         entry   = ring_buffer_event_data(event);
1548         entry->ip                       = ip;
1549         entry->parent_ip                = parent_ip;
1550
1551         trace_nowake_buffer_unlock_commit(buffer, event, flags, pc);
1552
1553  out:
1554         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
1555         ftrace_preempt_enable(resched);
1556 }
1557
1558 static struct ftrace_ops trace_ops __initdata  =
1559 {
1560         .func = function_test_events_call,
1561 };
1562
1563 static __init void event_trace_self_test_with_function(void)
1564 {
1565         register_ftrace_function(&trace_ops);
1566         pr_info("Running tests again, along with the function tracer\n");
1567         event_trace_self_tests();
1568         unregister_ftrace_function(&trace_ops);
1569 }
1570 #else
1571 static __init void event_trace_self_test_with_function(void)
1572 {
1573 }
1574 #endif
1575
1576 static __init int event_trace_self_tests_init(void)
1577 {
1578         if (!tracing_selftest_disabled) {
1579                 event_trace_self_tests();
1580                 event_trace_self_test_with_function();
1581         }
1582
1583         return 0;
1584 }
1585
1586 late_initcall(event_trace_self_tests_init);
1587
1588 #endif