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