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

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