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 <asm/setup.h>
  21
  22 #include "trace_output.h"
  23
  24 #undef TRACE_SYSTEM
  25 #define TRACE_SYSTEM "TRACE_SYSTEM"
  26
  27 DEFINE_MUTEX(event_mutex);
  28
  29 LIST_HEAD(ftrace_events);
  30
  31 int trace_define_field(struct ftrace_event_call *call, const char *type,
  32                        const char *name, int offset, int size, int is_signed,
  33                        int filter_type)
  34 {
  35         struct ftrace_event_field *field;
  36
  37         field = kzalloc(sizeof(*field), GFP_KERNEL);
  38         if (!field)
  39                 goto err;
  40
  41         field->name = kstrdup(name, GFP_KERNEL);
  42         if (!field->name)
  43                 goto err;
  44
  45         field->type = kstrdup(type, GFP_KERNEL);
  46         if (!field->type)
  47                 goto err;
  48
  49         if (filter_type == FILTER_OTHER)
  50                 field->filter_type = filter_assign_type(type);
  51         else
  52                 field->filter_type = filter_type;
  53
  54         field->offset = offset;
  55         field->size = size;
  56         field->is_signed = is_signed;
  57
  58         list_add(&field->link, &call->fields);
  59
  60         return 0;
  61
  62 err:
  63         if (field) {
  64                 kfree(field->name);
  65                 kfree(field->type);
  66         }
  67         kfree(field);
  68
  69         return -ENOMEM;
  70 }
  71 EXPORT_SYMBOL_GPL(trace_define_field);
  72
  73 #define __common_field(type, item)                                      \
  74         ret = trace_define_field(call, #type, "common_" #item,          \
  75                                  offsetof(typeof(ent), item),           \
  76                                  sizeof(ent.item),                      \
  77                                  is_signed_type(type), FILTER_OTHER);   \
  78         if (ret)                                                        \
  79                 return ret;
  80
  81 int trace_define_common_fields(struct ftrace_event_call *call)
  82 {
  83         int ret;
  84         struct trace_entry ent;
  85
  86         __common_field(unsigned short, type);
  87         __common_field(unsigned char, flags);
  88         __common_field(unsigned char, preempt_count);
  89         __common_field(int, pid);
  90         __common_field(int, lock_depth);
  91
  92         return ret;
  93 }
  94 EXPORT_SYMBOL_GPL(trace_define_common_fields);
  95
  96 void trace_destroy_fields(struct ftrace_event_call *call)
  97 {
  98         struct ftrace_event_field *field, *next;
  99
 100         list_for_each_entry_safe(field, next, &call->fields, link) {
 101                 list_del(&field->link);
 102                 kfree(field->type);
 103                 kfree(field->name);
 104                 kfree(field);
 105         }
 106 }
 107
 108 static void ftrace_event_enable_disable(struct ftrace_event_call *call,
 109                                         int enable)
 110 {
 111         switch (enable) {
 112         case 0:
 113                 if (call->enabled) {
 114                         call->enabled = 0;
 115                         tracing_stop_cmdline_record();
 116                         call->unregfunc(call);
 117                 }
 118                 break;
 119         case 1:
 120                 if (!call->enabled) {
 121                         call->enabled = 1;
 122                         tracing_start_cmdline_record();
 123                         call->regfunc(call);
 124                 }
 125                 break;
 126         }
 127 }
 128
 129 static void ftrace_clear_events(void)
 130 {
 131         struct ftrace_event_call *call;
 132
 133         mutex_lock(&event_mutex);
 134         list_for_each_entry(call, &ftrace_events, list) {
 135                 ftrace_event_enable_disable(call, 0);
 136         }
 137         mutex_unlock(&event_mutex);
 138 }
 139
 140 /*
 141  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 142  */
 143 static int __ftrace_set_clr_event(const char *match, const char *sub,
 144                                   const char *event, int set)
 145 {
 146         struct ftrace_event_call *call;
 147         int ret = -EINVAL;
 148
 149         mutex_lock(&event_mutex);
 150         list_for_each_entry(call, &ftrace_events, list) {
 151
 152                 if (!call->name || !call->regfunc)
 153                         continue;
 154
 155                 if (match &&
 156                     strcmp(match, call->name) != 0 &&
 157                     strcmp(match, call->system) != 0)
 158                         continue;
 159
 160                 if (sub && strcmp(sub, call->system) != 0)
 161                         continue;
 162
 163                 if (event && strcmp(event, call->name) != 0)
 164                         continue;
 165
 166                 ftrace_event_enable_disable(call, set);
 167
 168                 ret = 0;
 169         }
 170         mutex_unlock(&event_mutex);
 171
 172         return ret;
 173 }
 174
 175 static int ftrace_set_clr_event(char *buf, int set)
 176 {
 177         char *event = NULL, *sub = NULL, *match;
 178
 179         /*
 180          * The buf format can be <subsystem>:<event-name>
 181          *  *:<event-name> means any event by that name.
 182          *  :<event-name> is the same.
 183          *
 184          *  <subsystem>:* means all events in that subsystem
 185          *  <subsystem>: means the same.
 186          *
 187          *  <name> (no ':') means all events in a subsystem with
 188          *  the name <name> or any event that matches <name>
 189          */
 190
 191         match = strsep(&buf, ":");
 192         if (buf) {
 193                 sub = match;
 194                 event = buf;
 195                 match = NULL;
 196
 197                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 198                         sub = NULL;
 199                 if (!strlen(event) || strcmp(event, "*") == 0)
 200                         event = NULL;
 201         }
 202
 203         return __ftrace_set_clr_event(match, sub, event, set);
 204 }
 205
 206 /**
 207  * trace_set_clr_event - enable or disable an event
 208  * @system: system name to match (NULL for any system)
 209  * @event: event name to match (NULL for all events, within system)
 210  * @set: 1 to enable, 0 to disable
 211  *
 212  * This is a way for other parts of the kernel to enable or disable
 213  * event recording.
 214  *
 215  * Returns 0 on success, -EINVAL if the parameters do not match any
 216  * registered events.
 217  */
 218 int trace_set_clr_event(const char *system, const char *event, int set)
 219 {
 220         return __ftrace_set_clr_event(NULL, system, event, set);
 221 }
 222
 223 /* 128 should be much more than enough */
 224 #define EVENT_BUF_SIZE          127
 225
 226 static ssize_t
 227 ftrace_event_write(struct file *file, const char __user *ubuf,
 228                    size_t cnt, loff_t *ppos)
 229 {
 230         struct trace_parser parser;
 231         ssize_t read, ret;
 232
 233         if (!cnt)
 234                 return 0;
 235
 236         ret = tracing_update_buffers();
 237         if (ret < 0)
 238                 return ret;
 239
 240         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 241                 return -ENOMEM;
 242
 243         read = trace_get_user(&parser, ubuf, cnt, ppos);
 244
 245         if (read >= 0 && trace_parser_loaded((&parser))) {
 246                 int set = 1;
 247
 248                 if (*parser.buffer == '!')
 249                         set = 0;
 250
 251                 parser.buffer[parser.idx] = 0;
 252
 253                 ret = ftrace_set_clr_event(parser.buffer + !set, set);
 254                 if (ret)
 255                         goto out_put;
 256         }
 257
 258         ret = read;
 259
 260  out_put:
 261         trace_parser_put(&parser);
 262
 263         return ret;
 264 }
 265
 266 static void *
 267 t_next(struct seq_file *m, void *v, loff_t *pos)
 268 {
 269         struct ftrace_event_call *call = v;
 270
 271         (*pos)++;
 272
 273         list_for_each_entry_continue(call, &ftrace_events, list) {
 274                 /*
 275                  * The ftrace subsystem is for showing formats only.
 276                  * They can not be enabled or disabled via the event files.
 277                  */
 278                 if (call->regfunc)
 279                         return call;
 280         }
 281
 282         return NULL;
 283 }
 284
 285 static void *t_start(struct seq_file *m, loff_t *pos)
 286 {
 287         struct ftrace_event_call *call;
 288         loff_t l;
 289
 290         mutex_lock(&event_mutex);
 291
 292         call = list_entry(&ftrace_events, struct ftrace_event_call, list);
 293         for (l = 0; l <= *pos; ) {
 294                 call = t_next(m, call, &l);
 295                 if (!call)
 296                         break;
 297         }
 298         return call;
 299 }
 300
 301 static void *
 302 s_next(struct seq_file *m, void *v, loff_t *pos)
 303 {
 304         struct ftrace_event_call *call = v;
 305
 306         (*pos)++;
 307
 308         list_for_each_entry_continue(call, &ftrace_events, list) {
 309                 if (call->enabled)
 310                         return call;
 311         }
 312
 313         return NULL;
 314 }
 315
 316 static void *s_start(struct seq_file *m, loff_t *pos)
 317 {
 318         struct ftrace_event_call *call;
 319         loff_t l;
 320
 321         mutex_lock(&event_mutex);
 322
 323         call = list_entry(&ftrace_events, struct ftrace_event_call, list);
 324         for (l = 0; l <= *pos; ) {
 325                 call = s_next(m, call, &l);
 326                 if (!call)
 327                         break;
 328         }
 329         return call;
 330 }
 331
 332 static int t_show(struct seq_file *m, void *v)
 333 {
 334         struct ftrace_event_call *call = v;
 335
 336         if (strcmp(call->system, TRACE_SYSTEM) != 0)
 337                 seq_printf(m, "%s:", call->system);
 338         seq_printf(m, "%s\n", call->name);
 339
 340         return 0;
 341 }
 342
 343 static void t_stop(struct seq_file *m, void *p)
 344 {
 345         mutex_unlock(&event_mutex);
 346 }
 347
 348 static int
 349 ftrace_event_seq_open(struct inode *inode, struct file *file)
 350 {
 351         const struct seq_operations *seq_ops;
 352
 353         if ((file->f_mode & FMODE_WRITE) &&
 354             (file->f_flags & O_TRUNC))
 355                 ftrace_clear_events();
 356
 357         seq_ops = inode->i_private;
 358         return seq_open(file, seq_ops);
 359 }
 360
 361 static ssize_t
 362 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 363                   loff_t *ppos)
 364 {
 365         struct ftrace_event_call *call = filp->private_data;
 366         char *buf;
 367
 368         if (call->enabled)
 369                 buf = "1\n";
 370         else
 371                 buf = "0\n";
 372
 373         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 374 }
 375
 376 static ssize_t
 377 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 378                    loff_t *ppos)
 379 {
 380         struct ftrace_event_call *call = filp->private_data;
 381         char buf[64];
 382         unsigned long val;
 383         int ret;
 384
 385         if (cnt >= sizeof(buf))
 386                 return -EINVAL;
 387
 388         if (copy_from_user(&buf, ubuf, cnt))
 389                 return -EFAULT;
 390
 391         buf[cnt] = 0;
 392
 393         ret = strict_strtoul(buf, 10, &val);
 394         if (ret < 0)
 395                 return ret;
 396
 397         ret = tracing_update_buffers();
 398         if (ret < 0)
 399                 return ret;
 400
 401         switch (val) {
 402         case 0:
 403         case 1:
 404                 mutex_lock(&event_mutex);
 405                 ftrace_event_enable_disable(call, val);
 406                 mutex_unlock(&event_mutex);
 407                 break;
 408
 409         default:
 410                 return -EINVAL;
 411         }
 412
 413         *ppos += cnt;
 414
 415         return cnt;
 416 }
 417
 418 static ssize_t
 419 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 420                    loff_t *ppos)
 421 {
 422         const char set_to_char[4] = { '?', '0', '1', 'X' };
 423         const char *system = filp->private_data;
 424         struct ftrace_event_call *call;
 425         char buf[2];
 426         int set = 0;
 427         int ret;
 428
 429         mutex_lock(&event_mutex);
 430         list_for_each_entry(call, &ftrace_events, list) {
 431                 if (!call->name || !call->regfunc)
 432                         continue;
 433
 434                 if (system && strcmp(call->system, system) != 0)
 435                         continue;
 436
 437                 /*
 438                  * We need to find out if all the events are set
 439                  * or if all events or cleared, or if we have
 440                  * a mixture.
 441                  */
 442                 set |= (1 << !!call->enabled);
 443
 444                 /*
 445                  * If we have a mixture, no need to look further.
 446                  */
 447                 if (set == 3)
 448                         break;
 449         }
 450         mutex_unlock(&event_mutex);
 451
 452         buf[0] = set_to_char[set];
 453         buf[1] = '\n';
 454
 455         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 456
 457         return ret;
 458 }
 459
 460 static ssize_t
 461 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 462                     loff_t *ppos)
 463 {
 464         const char *system = filp->private_data;
 465         unsigned long val;
 466         char buf[64];
 467         ssize_t ret;
 468
 469         if (cnt >= sizeof(buf))
 470                 return -EINVAL;
 471
 472         if (copy_from_user(&buf, ubuf, cnt))
 473                 return -EFAULT;
 474
 475         buf[cnt] = 0;
 476
 477         ret = strict_strtoul(buf, 10, &val);
 478         if (ret < 0)
 479                 return ret;
 480
 481         ret = tracing_update_buffers();
 482         if (ret < 0)
 483                 return ret;
 484
 485         if (val != 0 && val != 1)
 486                 return -EINVAL;
 487
 488         ret = __ftrace_set_clr_event(NULL, system, NULL, val);
 489         if (ret)
 490                 goto out;
 491
 492         ret = cnt;
 493
 494 out:
 495         *ppos += cnt;
 496
 497         return ret;
 498 }
 499
 500 extern char *__bad_type_size(void);
 501
 502 #undef FIELD
 503 #define FIELD(type, name)                                               \
 504         sizeof(type) != sizeof(field.name) ? __bad_type_size() :        \
 505         #type, "common_" #name, offsetof(typeof(field), name),          \
 506                 sizeof(field.name), is_signed_type(type)
 507
 508 static int trace_write_header(struct trace_seq *s)
 509 {
 510         struct trace_entry field;
 511
 512         /* struct trace_entry */
 513         return trace_seq_printf(s,
 514                         "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\tsigned:%u;\n"
 515                         "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\tsigned:%u;\n"
 516                         "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\tsigned:%u;\n"
 517                         "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\tsigned:%u;\n"
 518                         "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\tsigned:%u;\n"
 519                         "\n",
 520                         FIELD(unsigned short, type),
 521                         FIELD(unsigned char, flags),
 522                         FIELD(unsigned char, preempt_count),
 523                         FIELD(int, pid),
 524                         FIELD(int, lock_depth));
 525 }
 526
 527 static ssize_t
 528 event_format_read(struct file *filp, char __user *ubuf, size_t cnt,
 529                   loff_t *ppos)
 530 {
 531         struct ftrace_event_call *call = filp->private_data;
 532         struct trace_seq *s;
 533         char *buf;
 534         int r;
 535
 536         if (*ppos)
 537                 return 0;
 538
 539         s = kmalloc(sizeof(*s), GFP_KERNEL);
 540         if (!s)
 541                 return -ENOMEM;
 542
 543         trace_seq_init(s);
 544
 545         /* If any of the first writes fail, so will the show_format. */
 546
 547         trace_seq_printf(s, "name: %s\n", call->name);
 548         trace_seq_printf(s, "ID: %d\n", call->id);
 549         trace_seq_printf(s, "format:\n");
 550         trace_write_header(s);
 551
 552         r = call->show_format(call, s);
 553         if (!r) {
 554                 /*
 555                  * ug!  The format output is bigger than a PAGE!!
 556                  */
 557                 buf = "FORMAT TOO BIG\n";
 558                 r = simple_read_from_buffer(ubuf, cnt, ppos,
 559                                               buf, strlen(buf));
 560                 goto out;
 561         }
 562
 563         r = simple_read_from_buffer(ubuf, cnt, ppos,
 564                                     s->buffer, s->len);
 565  out:
 566         kfree(s);
 567         return r;
 568 }
 569
 570 static ssize_t
 571 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 572 {
 573         struct ftrace_event_call *call = filp->private_data;
 574         struct trace_seq *s;
 575         int r;
 576
 577         if (*ppos)
 578                 return 0;
 579
 580         s = kmalloc(sizeof(*s), GFP_KERNEL);
 581         if (!s)
 582                 return -ENOMEM;
 583
 584         trace_seq_init(s);
 585         trace_seq_printf(s, "%d\n", call->id);
 586
 587         r = simple_read_from_buffer(ubuf, cnt, ppos,
 588                                     s->buffer, s->len);
 589         kfree(s);
 590         return r;
 591 }
 592
 593 static ssize_t
 594 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 595                   loff_t *ppos)
 596 {
 597         struct ftrace_event_call *call = filp->private_data;
 598         struct trace_seq *s;
 599         int r;
 600
 601         if (*ppos)
 602                 return 0;
 603
 604         s = kmalloc(sizeof(*s), GFP_KERNEL);
 605         if (!s)
 606                 return -ENOMEM;
 607
 608         trace_seq_init(s);
 609
 610         print_event_filter(call, s);
 611         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 612
 613         kfree(s);
 614
 615         return r;
 616 }
 617
 618 static ssize_t
 619 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
 620                    loff_t *ppos)
 621 {
 622         struct ftrace_event_call *call = filp->private_data;
 623         char *buf;
 624         int err;
 625
 626         if (cnt >= PAGE_SIZE)
 627                 return -EINVAL;
 628
 629         buf = (char *)__get_free_page(GFP_TEMPORARY);
 630         if (!buf)
 631                 return -ENOMEM;
 632
 633         if (copy_from_user(buf, ubuf, cnt)) {
 634                 free_page((unsigned long) buf);
 635                 return -EFAULT;
 636         }
 637         buf[cnt] = '\0';
 638
 639         err = apply_event_filter(call, buf);
 640         free_page((unsigned long) buf);
 641         if (err < 0)
 642                 return err;
 643
 644         *ppos += cnt;
 645
 646         return cnt;
 647 }
 648
 649 static ssize_t
 650 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 651                       loff_t *ppos)
 652 {
 653         struct event_subsystem *system = filp->private_data;
 654         struct trace_seq *s;
 655         int r;
 656
 657         if (*ppos)
 658                 return 0;
 659
 660         s = kmalloc(sizeof(*s), GFP_KERNEL);
 661         if (!s)
 662                 return -ENOMEM;
 663
 664         trace_seq_init(s);
 665
 666         print_subsystem_event_filter(system, s);
 667         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 668
 669         kfree(s);
 670
 671         return r;
 672 }
 673
 674 static ssize_t
 675 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
 676                        loff_t *ppos)
 677 {
 678         struct event_subsystem *system = filp->private_data;
 679         char *buf;
 680         int err;
 681
 682         if (cnt >= PAGE_SIZE)
 683                 return -EINVAL;
 684
 685         buf = (char *)__get_free_page(GFP_TEMPORARY);
 686         if (!buf)
 687                 return -ENOMEM;
 688
 689         if (copy_from_user(buf, ubuf, cnt)) {
 690                 free_page((unsigned long) buf);
 691                 return -EFAULT;
 692         }
 693         buf[cnt] = '\0';
 694
 695         err = apply_subsystem_event_filter(system, buf);
 696         free_page((unsigned long) buf);
 697         if (err < 0)
 698                 return err;
 699
 700         *ppos += cnt;
 701
 702         return cnt;
 703 }
 704
 705 static ssize_t
 706 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 707 {
 708         int (*func)(struct trace_seq *s) = filp->private_data;
 709         struct trace_seq *s;
 710         int r;
 711
 712         if (*ppos)
 713                 return 0;
 714
 715         s = kmalloc(sizeof(*s), GFP_KERNEL);
 716         if (!s)
 717                 return -ENOMEM;
 718
 719         trace_seq_init(s);
 720
 721         func(s);
 722         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 723
 724         kfree(s);
 725
 726         return r;
 727 }
 728
 729 static const struct seq_operations show_event_seq_ops = {
 730         .start = t_start,
 731         .next = t_next,
 732         .show = t_show,
 733         .stop = t_stop,
 734 };
 735
 736 static const struct seq_operations show_set_event_seq_ops = {
 737         .start = s_start,
 738         .next = s_next,
 739         .show = t_show,
 740         .stop = t_stop,
 741 };
 742
 743 static const struct file_operations ftrace_avail_fops = {
 744         .open = ftrace_event_seq_open,
 745         .read = seq_read,
 746         .llseek = seq_lseek,
 747         .release = seq_release,
 748 };
 749
 750 static const struct file_operations ftrace_set_event_fops = {
 751         .open = ftrace_event_seq_open,
 752         .read = seq_read,
 753         .write = ftrace_event_write,
 754         .llseek = seq_lseek,
 755         .release = seq_release,
 756 };
 757
 758 static const struct file_operations ftrace_enable_fops = {
 759         .open = tracing_open_generic,
 760         .read = event_enable_read,
 761         .write = event_enable_write,
 762 };
 763
 764 static const struct file_operations ftrace_event_format_fops = {
 765         .open = tracing_open_generic,
 766         .read = event_format_read,
 767 };
 768
 769 static const struct file_operations ftrace_event_id_fops = {
 770         .open = tracing_open_generic,
 771         .read = event_id_read,
 772 };
 773
 774 static const struct file_operations ftrace_event_filter_fops = {
 775         .open = tracing_open_generic,
 776         .read = event_filter_read,
 777         .write = event_filter_write,
 778 };
 779
 780 static const struct file_operations ftrace_subsystem_filter_fops = {
 781         .open = tracing_open_generic,
 782         .read = subsystem_filter_read,
 783         .write = subsystem_filter_write,
 784 };
 785
 786 static const struct file_operations ftrace_system_enable_fops = {
 787         .open = tracing_open_generic,
 788         .read = system_enable_read,
 789         .write = system_enable_write,
 790 };
 791
 792 static const struct file_operations ftrace_show_header_fops = {
 793         .open = tracing_open_generic,
 794         .read = show_header,
 795 };
 796
 797 static struct dentry *event_trace_events_dir(void)
 798 {
 799         static struct dentry *d_tracer;
 800         static struct dentry *d_events;
 801
 802         if (d_events)
 803                 return d_events;
 804
 805         d_tracer = tracing_init_dentry();
 806         if (!d_tracer)
 807                 return NULL;
 808
 809         d_events = debugfs_create_dir("events", d_tracer);
 810         if (!d_events)
 811                 pr_warning("Could not create debugfs "
 812                            "'events' directory\n");
 813
 814         return d_events;
 815 }
 816
 817 static LIST_HEAD(event_subsystems);
 818
 819 static struct dentry *
 820 event_subsystem_dir(const char *name, struct dentry *d_events)
 821 {
 822         struct event_subsystem *system;
 823         struct dentry *entry;
 824
 825         /* First see if we did not already create this dir */
 826         list_for_each_entry(system, &event_subsystems, list) {
 827                 if (strcmp(system->name, name) == 0) {
 828                         system->nr_events++;
 829                         return system->entry;
 830                 }
 831         }
 832
 833         /* need to create new entry */
 834         system = kmalloc(sizeof(*system), GFP_KERNEL);
 835         if (!system) {
 836                 pr_warning("No memory to create event subsystem %s\n",
 837                            name);
 838                 return d_events;
 839         }
 840
 841         system->entry = debugfs_create_dir(name, d_events);
 842         if (!system->entry) {
 843                 pr_warning("Could not create event subsystem %s\n",
 844                            name);
 845                 kfree(system);
 846                 return d_events;
 847         }
 848
 849         system->nr_events = 1;
 850         system->name = kstrdup(name, GFP_KERNEL);
 851         if (!system->name) {
 852                 debugfs_remove(system->entry);
 853                 kfree(system);
 854                 return d_events;
 855         }
 856
 857         list_add(&system->list, &event_subsystems);
 858
 859         system->filter = NULL;
 860
 861         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
 862         if (!system->filter) {
 863                 pr_warning("Could not allocate filter for subsystem "
 864                            "'%s'\n", name);
 865                 return system->entry;
 866         }
 867
 868         entry = debugfs_create_file("filter", 0644, system->entry, system,
 869                                     &ftrace_subsystem_filter_fops);
 870         if (!entry) {
 871                 kfree(system->filter);
 872                 system->filter = NULL;
 873                 pr_warning("Could not create debugfs "
 874                            "'%s/filter' entry\n", name);
 875         }
 876
 877         trace_create_file("enable", 0644, system->entry,
 878                           (void *)system->name,
 879                           &ftrace_system_enable_fops);
 880
 881         return system->entry;
 882 }
 883
 884 static int
 885 event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
 886                  const struct file_operations *id,
 887                  const struct file_operations *enable,
 888                  const struct file_operations *filter,
 889                  const struct file_operations *format)
 890 {
 891         int ret;
 892
 893         /*
 894          * If the trace point header did not define TRACE_SYSTEM
 895          * then the system would be called "TRACE_SYSTEM".
 896          */
 897         if (strcmp(call->system, TRACE_SYSTEM) != 0)
 898                 d_events = event_subsystem_dir(call->system, d_events);
 899
 900         call->dir = debugfs_create_dir(call->name, d_events);
 901         if (!call->dir) {
 902                 pr_warning("Could not create debugfs "
 903                            "'%s' directory\n", call->name);
 904                 return -1;
 905         }
 906
 907         if (call->regfunc)
 908                 trace_create_file("enable", 0644, call->dir, call,
 909                                   enable);
 910
 911         if (call->id && call->profile_enable)
 912                 trace_create_file("id", 0444, call->dir, call,
 913                                   id);
 914
 915         if (call->define_fields) {
 916                 ret = call->define_fields(call);
 917                 if (ret < 0) {
 918                         pr_warning("Could not initialize trace point"
 919                                    " events/%s\n", call->name);
 920                         return ret;
 921                 }
 922                 trace_create_file("filter", 0644, call->dir, call,
 923                                   filter);
 924         }
 925
 926         /* A trace may not want to export its format */
 927         if (!call->show_format)
 928                 return 0;
 929
 930         trace_create_file("format", 0444, call->dir, call,
 931                           format);
 932
 933         return 0;
 934 }
 935
 936 static int __trace_add_event_call(struct ftrace_event_call *call)
 937 {
 938         struct dentry *d_events;
 939         int ret;
 940
 941         if (!call->name)
 942                 return -EINVAL;
 943
 944         if (call->raw_init) {
 945                 ret = call->raw_init(call);
 946                 if (ret < 0) {
 947                         if (ret != -ENOSYS)
 948                                 pr_warning("Could not initialize trace "
 949                                 "events/%s\n", call->name);
 950                         return ret;
 951                 }
 952         }
 953
 954         d_events = event_trace_events_dir();
 955         if (!d_events)
 956                 return -ENOENT;
 957
 958         ret = event_create_dir(call, d_events, &ftrace_event_id_fops,
 959                                 &ftrace_enable_fops, &ftrace_event_filter_fops,
 960                                 &ftrace_event_format_fops);
 961         if (!ret)
 962                 list_add(&call->list, &ftrace_events);
 963
 964         return ret;
 965 }
 966
 967 /* Add an additional event_call dynamically */
 968 int trace_add_event_call(struct ftrace_event_call *call)
 969 {
 970         int ret;
 971         mutex_lock(&event_mutex);
 972         ret = __trace_add_event_call(call);
 973         mutex_unlock(&event_mutex);
 974         return ret;
 975 }
 976
 977 static void remove_subsystem_dir(const char *name)
 978 {
 979         struct event_subsystem *system;
 980
 981         if (strcmp(name, TRACE_SYSTEM) == 0)
 982                 return;
 983
 984         list_for_each_entry(system, &event_subsystems, list) {
 985                 if (strcmp(system->name, name) == 0) {
 986                         if (!--system->nr_events) {
 987                                 struct event_filter *filter = system->filter;
 988
 989                                 debugfs_remove_recursive(system->entry);
 990                                 list_del(&system->list);
 991                                 if (filter) {
 992                                         kfree(filter->filter_string);
 993                                         kfree(filter);
 994                                 }
 995                                 kfree(system->name);
 996                                 kfree(system);
 997                         }
 998                         break;
 999                 }
1000         }
1001 }
1002
1003 /*
1004  * Must be called under locking both of event_mutex and trace_event_mutex.
1005  */
1006 static void __trace_remove_event_call(struct ftrace_event_call *call)
1007 {
1008         ftrace_event_enable_disable(call, 0);
1009         if (call->event)
1010                 __unregister_ftrace_event(call->event);
1011         debugfs_remove_recursive(call->dir);
1012         list_del(&call->list);
1013         trace_destroy_fields(call);
1014         destroy_preds(call);
1015         remove_subsystem_dir(call->system);
1016 }
1017
1018 /* Remove an event_call */
1019 void trace_remove_event_call(struct ftrace_event_call *call)
1020 {
1021         mutex_lock(&event_mutex);
1022         down_write(&trace_event_mutex);
1023         __trace_remove_event_call(call);
1024         up_write(&trace_event_mutex);
1025         mutex_unlock(&event_mutex);
1026 }
1027
1028 #define for_each_event(event, start, end)                       \
1029         for (event = start;                                     \
1030              (unsigned long)event < (unsigned long)end;         \
1031              event++)
1032
1033 #ifdef CONFIG_MODULES
1034
1035 static LIST_HEAD(ftrace_module_file_list);
1036
1037 /*
1038  * Modules must own their file_operations to keep up with
1039  * reference counting.
1040  */
1041 struct ftrace_module_file_ops {
1042         struct list_head                list;
1043         struct module                   *mod;
1044         struct file_operations          id;
1045         struct file_operations          enable;
1046         struct file_operations          format;
1047         struct file_operations          filter;
1048 };
1049
1050 static struct ftrace_module_file_ops *
1051 trace_create_file_ops(struct module *mod)
1052 {
1053         struct ftrace_module_file_ops *file_ops;
1054
1055         /*
1056          * This is a bit of a PITA. To allow for correct reference
1057          * counting, modules must "own" their file_operations.
1058          * To do this, we allocate the file operations that will be
1059          * used in the event directory.
1060          */
1061
1062         file_ops = kmalloc(sizeof(*file_ops), GFP_KERNEL);
1063         if (!file_ops)
1064                 return NULL;
1065
1066         file_ops->mod = mod;
1067
1068         file_ops->id = ftrace_event_id_fops;
1069         file_ops->id.owner = mod;
1070
1071         file_ops->enable = ftrace_enable_fops;
1072         file_ops->enable.owner = mod;
1073
1074         file_ops->filter = ftrace_event_filter_fops;
1075         file_ops->filter.owner = mod;
1076
1077         file_ops->format = ftrace_event_format_fops;
1078         file_ops->format.owner = mod;
1079
1080         list_add(&file_ops->list, &ftrace_module_file_list);
1081
1082         return file_ops;
1083 }
1084
1085 static void trace_module_add_events(struct module *mod)
1086 {
1087         struct ftrace_module_file_ops *file_ops = NULL;
1088         struct ftrace_event_call *call, *start, *end;
1089         struct dentry *d_events;
1090         int ret;
1091
1092         start = mod->trace_events;
1093         end = mod->trace_events + mod->num_trace_events;
1094
1095         if (start == end)
1096                 return;
1097
1098         d_events = event_trace_events_dir();
1099         if (!d_events)
1100                 return;
1101
1102         for_each_event(call, start, end) {
1103                 /* The linker may leave blanks */
1104                 if (!call->name)
1105                         continue;
1106                 if (call->raw_init) {
1107                         ret = call->raw_init(call);
1108                         if (ret < 0) {
1109                                 if (ret != -ENOSYS)
1110                                         pr_warning("Could not initialize trace "
1111                                         "point events/%s\n", call->name);
1112                                 continue;
1113                         }
1114                 }
1115                 /*
1116                  * This module has events, create file ops for this module
1117                  * if not already done.
1118                  */
1119                 if (!file_ops) {
1120                         file_ops = trace_create_file_ops(mod);
1121                         if (!file_ops)
1122                                 return;
1123                 }
1124                 call->mod = mod;
1125                 ret = event_create_dir(call, d_events,
1126                                        &file_ops->id, &file_ops->enable,
1127                                        &file_ops->filter, &file_ops->format);
1128                 if (!ret)
1129                         list_add(&call->list, &ftrace_events);
1130         }
1131 }
1132
1133 static void trace_module_remove_events(struct module *mod)
1134 {
1135         struct ftrace_module_file_ops *file_ops;
1136         struct ftrace_event_call *call, *p;
1137         bool found = false;
1138
1139         down_write(&trace_event_mutex);
1140         list_for_each_entry_safe(call, p, &ftrace_events, list) {
1141                 if (call->mod == mod) {
1142                         found = true;
1143                         __trace_remove_event_call(call);
1144                 }
1145         }
1146
1147         /* Now free the file_operations */
1148         list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1149                 if (file_ops->mod == mod)
1150                         break;
1151         }
1152         if (&file_ops->list != &ftrace_module_file_list) {
1153                 list_del(&file_ops->list);
1154                 kfree(file_ops);
1155         }
1156
1157         /*
1158          * It is safest to reset the ring buffer if the module being unloaded
1159          * registered any events.
1160          */
1161         if (found)
1162                 tracing_reset_current_online_cpus();
1163         up_write(&trace_event_mutex);
1164 }
1165
1166 static int trace_module_notify(struct notifier_block *self,
1167                                unsigned long val, void *data)
1168 {
1169         struct module *mod = data;
1170
1171         mutex_lock(&event_mutex);
1172         switch (val) {
1173         case MODULE_STATE_COMING:
1174                 trace_module_add_events(mod);
1175                 break;
1176         case MODULE_STATE_GOING:
1177                 trace_module_remove_events(mod);
1178                 break;
1179         }
1180         mutex_unlock(&event_mutex);
1181
1182         return 0;
1183 }
1184 #else
1185 static int trace_module_notify(struct notifier_block *self,
1186                                unsigned long val, void *data)
1187 {
1188         return 0;
1189 }
1190 #endif /* CONFIG_MODULES */
1191
1192 static struct notifier_block trace_module_nb = {
1193         .notifier_call = trace_module_notify,
1194         .priority = 0,
1195 };
1196
1197 extern struct ftrace_event_call __start_ftrace_events[];
1198 extern struct ftrace_event_call __stop_ftrace_events[];
1199
1200 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
1201
1202 static __init int setup_trace_event(char *str)
1203 {
1204         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
1205         ring_buffer_expanded = 1;
1206         tracing_selftest_disabled = 1;
1207
1208         return 1;
1209 }
1210 __setup("trace_event=", setup_trace_event);
1211
1212 static __init int event_trace_init(void)
1213 {
1214         struct ftrace_event_call *call;
1215         struct dentry *d_tracer;
1216         struct dentry *entry;
1217         struct dentry *d_events;
1218         int ret;
1219         char *buf = bootup_event_buf;
1220         char *token;
1221
1222         d_tracer = tracing_init_dentry();
1223         if (!d_tracer)
1224                 return 0;
1225
1226         entry = debugfs_create_file("available_events", 0444, d_tracer,
1227                                     (void *)&show_event_seq_ops,
1228                                     &ftrace_avail_fops);
1229         if (!entry)
1230                 pr_warning("Could not create debugfs "
1231                            "'available_events' entry\n");
1232
1233         entry = debugfs_create_file("set_event", 0644, d_tracer,
1234                                     (void *)&show_set_event_seq_ops,
1235                                     &ftrace_set_event_fops);
1236         if (!entry)
1237                 pr_warning("Could not create debugfs "
1238                            "'set_event' entry\n");
1239
1240         d_events = event_trace_events_dir();
1241         if (!d_events)
1242                 return 0;
1243
1244         /* ring buffer internal formats */
1245         trace_create_file("header_page", 0444, d_events,
1246                           ring_buffer_print_page_header,
1247                           &ftrace_show_header_fops);
1248
1249         trace_create_file("header_event", 0444, d_events,
1250                           ring_buffer_print_entry_header,
1251                           &ftrace_show_header_fops);
1252
1253         trace_create_file("enable", 0644, d_events,
1254                           NULL, &ftrace_system_enable_fops);
1255
1256         for_each_event(call, __start_ftrace_events, __stop_ftrace_events) {
1257                 /* The linker may leave blanks */
1258                 if (!call->name)
1259                         continue;
1260                 if (call->raw_init) {
1261                         ret = call->raw_init(call);
1262                         if (ret < 0) {
1263                                 if (ret != -ENOSYS)
1264                                         pr_warning("Could not initialize trace "
1265                                         "point events/%s\n", call->name);
1266                                 continue;
1267                         }
1268                 }
1269                 ret = event_create_dir(call, d_events, &ftrace_event_id_fops,
1270                                        &ftrace_enable_fops,
1271                                        &ftrace_event_filter_fops,
1272                                        &ftrace_event_format_fops);
1273                 if (!ret)
1274                         list_add(&call->list, &ftrace_events);
1275         }
1276
1277         while (true) {
1278                 token = strsep(&buf, ",");
1279
1280                 if (!token)
1281                         break;
1282                 if (!*token)
1283                         continue;
1284
1285                 ret = ftrace_set_clr_event(token, 1);
1286                 if (ret)
1287                         pr_warning("Failed to enable trace event: %s\n", token);
1288         }
1289
1290         ret = register_module_notifier(&trace_module_nb);
1291         if (ret)
1292                 pr_warning("Failed to register trace events module notifier\n");
1293
1294         return 0;
1295 }
1296 fs_initcall(event_trace_init);
1297
1298 #ifdef CONFIG_FTRACE_STARTUP_TEST
1299
1300 static DEFINE_SPINLOCK(test_spinlock);
1301 static DEFINE_SPINLOCK(test_spinlock_irq);
1302 static DEFINE_MUTEX(test_mutex);
1303
1304 static __init void test_work(struct work_struct *dummy)
1305 {
1306         spin_lock(&test_spinlock);
1307         spin_lock_irq(&test_spinlock_irq);
1308         udelay(1);
1309         spin_unlock_irq(&test_spinlock_irq);
1310         spin_unlock(&test_spinlock);
1311
1312         mutex_lock(&test_mutex);
1313         msleep(1);
1314         mutex_unlock(&test_mutex);
1315 }
1316
1317 static __init int event_test_thread(void *unused)
1318 {
1319         void *test_malloc;
1320
1321         test_malloc = kmalloc(1234, GFP_KERNEL);
1322         if (!test_malloc)
1323                 pr_info("failed to kmalloc\n");
1324
1325         schedule_on_each_cpu(test_work);
1326
1327         kfree(test_malloc);
1328
1329         set_current_state(TASK_INTERRUPTIBLE);
1330         while (!kthread_should_stop())
1331                 schedule();
1332
1333         return 0;
1334 }
1335
1336 /*
1337  * Do various things that may trigger events.
1338  */
1339 static __init void event_test_stuff(void)
1340 {
1341         struct task_struct *test_thread;
1342
1343         test_thread = kthread_run(event_test_thread, NULL, "test-events");
1344         msleep(1);
1345         kthread_stop(test_thread);
1346 }
1347
1348 /*
1349  * For every trace event defined, we will test each trace point separately,
1350  * and then by groups, and finally all trace points.
1351  */
1352 static __init void event_trace_self_tests(void)
1353 {
1354         struct ftrace_event_call *call;
1355         struct event_subsystem *system;
1356         int ret;
1357
1358         pr_info("Running tests on trace events:\n");
1359
1360         list_for_each_entry(call, &ftrace_events, list) {
1361
1362                 /* Only test those that have a regfunc */
1363                 if (!call->regfunc)
1364                         continue;
1365
1366 /*
1367  * Testing syscall events here is pretty useless, but
1368  * we still do it if configured. But this is time consuming.
1369  * What we really need is a user thread to perform the
1370  * syscalls as we test.
1371  */
1372 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
1373                 if (call->system &&
1374                     strcmp(call->system, "syscalls") == 0)
1375                         continue;
1376 #endif
1377
1378                 pr_info("Testing event %s: ", call->name);
1379
1380                 /*
1381                  * If an event is already enabled, someone is using
1382                  * it and the self test should not be on.
1383                  */
1384                 if (call->enabled) {
1385                         pr_warning("Enabled event during self test!\n");
1386                         WARN_ON_ONCE(1);
1387                         continue;
1388                 }
1389
1390                 ftrace_event_enable_disable(call, 1);
1391                 event_test_stuff();
1392                 ftrace_event_enable_disable(call, 0);
1393
1394                 pr_cont("OK\n");
1395         }
1396
1397         /* Now test at the sub system level */
1398
1399         pr_info("Running tests on trace event systems:\n");
1400
1401         list_for_each_entry(system, &event_subsystems, list) {
1402
1403                 /* the ftrace system is special, skip it */
1404                 if (strcmp(system->name, "ftrace") == 0)
1405                         continue;
1406
1407                 pr_info("Testing event system %s: ", system->name);
1408
1409                 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 1);
1410                 if (WARN_ON_ONCE(ret)) {
1411                         pr_warning("error enabling system %s\n",
1412                                    system->name);
1413                         continue;
1414                 }
1415
1416                 event_test_stuff();
1417
1418                 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 0);
1419                 if (WARN_ON_ONCE(ret))
1420                         pr_warning("error disabling system %s\n",
1421                                    system->name);
1422
1423                 pr_cont("OK\n");
1424         }
1425
1426         /* Test with all events enabled */
1427
1428         pr_info("Running tests on all trace events:\n");
1429         pr_info("Testing all events: ");
1430
1431         ret = __ftrace_set_clr_event(NULL, NULL, NULL, 1);
1432         if (WARN_ON_ONCE(ret)) {
1433                 pr_warning("error enabling all events\n");
1434                 return;
1435         }
1436
1437         event_test_stuff();
1438
1439         /* reset sysname */
1440         ret = __ftrace_set_clr_event(NULL, NULL, NULL, 0);
1441         if (WARN_ON_ONCE(ret)) {
1442                 pr_warning("error disabling all events\n");
1443                 return;
1444         }
1445
1446         pr_cont("OK\n");
1447 }
1448
1449 #ifdef CONFIG_FUNCTION_TRACER
1450
1451 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
1452
1453 static void
1454 function_test_events_call(unsigned long ip, unsigned long parent_ip)
1455 {
1456         struct ring_buffer_event *event;
1457         struct ring_buffer *buffer;
1458         struct ftrace_entry *entry;
1459         unsigned long flags;
1460         long disabled;
1461         int resched;
1462         int cpu;
1463         int pc;
1464
1465         pc = preempt_count();
1466         resched = ftrace_preempt_disable();
1467         cpu = raw_smp_processor_id();
1468         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
1469
1470         if (disabled != 1)
1471                 goto out;
1472
1473         local_save_flags(flags);
1474
1475         event = trace_current_buffer_lock_reserve(&buffer,
1476                                                   TRACE_FN, sizeof(*entry),
1477                                                   flags, pc);
1478         if (!event)
1479                 goto out;
1480         entry   = ring_buffer_event_data(event);
1481         entry->ip                       = ip;
1482         entry->parent_ip                = parent_ip;
1483
1484         trace_nowake_buffer_unlock_commit(buffer, event, flags, pc);
1485
1486  out:
1487         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
1488         ftrace_preempt_enable(resched);
1489 }
1490
1491 static struct ftrace_ops trace_ops __initdata  =
1492 {
1493         .func = function_test_events_call,
1494 };
1495
1496 static __init void event_trace_self_test_with_function(void)
1497 {
1498         register_ftrace_function(&trace_ops);
1499         pr_info("Running tests again, along with the function tracer\n");
1500         event_trace_self_tests();
1501         unregister_ftrace_function(&trace_ops);
1502 }
1503 #else
1504 static __init void event_trace_self_test_with_function(void)
1505 {
1506 }
1507 #endif
1508
1509 static __init int event_trace_self_tests_init(void)
1510 {
1511         if (!tracing_selftest_disabled) {
1512                 event_trace_self_tests();
1513                 event_trace_self_test_with_function();
1514         }
1515
1516         return 0;
1517 }
1518
1519 late_initcall(event_trace_self_tests_init);
1520
1521 #endif