kernel/trace/trace_events.c

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