git.oblomov.eu Git - 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/debugfs.h>
  12 #include <linux/uaccess.h>
  13 #include <linux/module.h>
  14 #include <linux/ctype.h>
  15
  16 #include "trace_output.h"
  17
  18 #define TRACE_SYSTEM "TRACE_SYSTEM"
  19
  20 static DEFINE_MUTEX(event_mutex);
  21
  22 static void ftrace_clear_events(void)
  23 {
  24         struct ftrace_event_call *call = (void *)__start_ftrace_events;
  25
  26
  27         while ((unsigned long)call < (unsigned long)__stop_ftrace_events) {
  28
  29                 if (call->enabled) {
  30                         call->enabled = 0;
  31                         call->unregfunc();
  32                 }
  33                 call++;
  34         }
  35 }
  36
  37 static void ftrace_event_enable_disable(struct ftrace_event_call *call,
  38                                         int enable)
  39 {
  40
  41         switch (enable) {
  42         case 0:
  43                 if (call->enabled) {
  44                         call->enabled = 0;
  45                         call->unregfunc();
  46                 }
  47                 break;
  48         case 1:
  49                 if (!call->enabled) {
  50                         call->enabled = 1;
  51                         call->regfunc();
  52                 }
  53                 break;
  54         }
  55 }
  56
  57 static int ftrace_set_clr_event(char *buf, int set)
  58 {
  59         struct ftrace_event_call *call = __start_ftrace_events;
  60         char *event = NULL, *sub = NULL, *match;
  61         int ret = -EINVAL;
  62
  63         /*
  64          * The buf format can be <subsystem>:<event-name>
  65          *  *:<event-name> means any event by that name.
  66          *  :<event-name> is the same.
  67          *
  68          *  <subsystem>:* means all events in that subsystem
  69          *  <subsystem>: means the same.
  70          *
  71          *  <name> (no ':') means all events in a subsystem with
  72          *  the name <name> or any event that matches <name>
  73          */
  74
  75         match = strsep(&buf, ":");
  76         if (buf) {
  77                 sub = match;
  78                 event = buf;
  79                 match = NULL;
  80
  81                 if (!strlen(sub) || strcmp(sub, "*") == 0)
  82                         sub = NULL;
  83                 if (!strlen(event) || strcmp(event, "*") == 0)
  84                         event = NULL;
  85         }
  86
  87         mutex_lock(&event_mutex);
  88         for_each_event(call) {
  89
  90                 if (!call->name || !call->regfunc)
  91                         continue;
  92
  93                 if (match &&
  94                     strcmp(match, call->name) != 0 &&
  95                     strcmp(match, call->system) != 0)
  96                         continue;
  97
  98                 if (sub && strcmp(sub, call->system) != 0)
  99                         continue;
 100
 101                 if (event && strcmp(event, call->name) != 0)
 102                         continue;
 103
 104                 ftrace_event_enable_disable(call, set);
 105
 106                 ret = 0;
 107         }
 108         mutex_unlock(&event_mutex);
 109
 110         return ret;
 111 }
 112
 113 /* 128 should be much more than enough */
 114 #define EVENT_BUF_SIZE          127
 115
 116 static ssize_t
 117 ftrace_event_write(struct file *file, const char __user *ubuf,
 118                    size_t cnt, loff_t *ppos)
 119 {
 120         size_t read = 0;
 121         int i, set = 1;
 122         ssize_t ret;
 123         char *buf;
 124         char ch;
 125
 126         if (!cnt || cnt < 0)
 127                 return 0;
 128
 129         ret = tracing_update_buffers();
 130         if (ret < 0)
 131                 return ret;
 132
 133         ret = get_user(ch, ubuf++);
 134         if (ret)
 135                 return ret;
 136         read++;
 137         cnt--;
 138
 139         /* skip white space */
 140         while (cnt && isspace(ch)) {
 141                 ret = get_user(ch, ubuf++);
 142                 if (ret)
 143                         return ret;
 144                 read++;
 145                 cnt--;
 146         }
 147
 148         /* Only white space found? */
 149         if (isspace(ch)) {
 150                 file->f_pos += read;
 151                 ret = read;
 152                 return ret;
 153         }
 154
 155         buf = kmalloc(EVENT_BUF_SIZE+1, GFP_KERNEL);
 156         if (!buf)
 157                 return -ENOMEM;
 158
 159         if (cnt > EVENT_BUF_SIZE)
 160                 cnt = EVENT_BUF_SIZE;
 161
 162         i = 0;
 163         while (cnt && !isspace(ch)) {
 164                 if (!i && ch == '!')
 165                         set = 0;
 166                 else
 167                         buf[i++] = ch;
 168
 169                 ret = get_user(ch, ubuf++);
 170                 if (ret)
 171                         goto out_free;
 172                 read++;
 173                 cnt--;
 174         }
 175         buf[i] = 0;
 176
 177         file->f_pos += read;
 178
 179         ret = ftrace_set_clr_event(buf, set);
 180         if (ret)
 181                 goto out_free;
 182
 183         ret = read;
 184
 185  out_free:
 186         kfree(buf);
 187
 188         return ret;
 189 }
 190
 191 static void *
 192 t_next(struct seq_file *m, void *v, loff_t *pos)
 193 {
 194         struct ftrace_event_call *call = m->private;
 195         struct ftrace_event_call *next = call;
 196
 197         (*pos)++;
 198
 199         for (;;) {
 200                 if ((unsigned long)call >= (unsigned long)__stop_ftrace_events)
 201                         return NULL;
 202
 203                 /*
 204                  * The ftrace subsystem is for showing formats only.
 205                  * They can not be enabled or disabled via the event files.
 206                  */
 207                 if (call->regfunc)
 208                         break;
 209
 210                 call++;
 211                 next = call;
 212         }
 213
 214         m->private = ++next;
 215
 216         return call;
 217 }
 218
 219 static void *t_start(struct seq_file *m, loff_t *pos)
 220 {
 221         return t_next(m, NULL, pos);
 222 }
 223
 224 static void *
 225 s_next(struct seq_file *m, void *v, loff_t *pos)
 226 {
 227         struct ftrace_event_call *call = m->private;
 228         struct ftrace_event_call *next;
 229
 230         (*pos)++;
 231
 232  retry:
 233         if ((unsigned long)call >= (unsigned long)__stop_ftrace_events)
 234                 return NULL;
 235
 236         if (!call->enabled) {
 237                 call++;
 238                 goto retry;
 239         }
 240
 241         next = call;
 242         m->private = ++next;
 243
 244         return call;
 245 }
 246
 247 static void *s_start(struct seq_file *m, loff_t *pos)
 248 {
 249         return s_next(m, NULL, pos);
 250 }
 251
 252 static int t_show(struct seq_file *m, void *v)
 253 {
 254         struct ftrace_event_call *call = v;
 255
 256         if (strcmp(call->system, TRACE_SYSTEM) != 0)
 257                 seq_printf(m, "%s:", call->system);
 258         seq_printf(m, "%s\n", call->name);
 259
 260         return 0;
 261 }
 262
 263 static void t_stop(struct seq_file *m, void *p)
 264 {
 265 }
 266
 267 static int
 268 ftrace_event_seq_open(struct inode *inode, struct file *file)
 269 {
 270         int ret;
 271         const struct seq_operations *seq_ops;
 272
 273         if ((file->f_mode & FMODE_WRITE) &&
 274             !(file->f_flags & O_APPEND))
 275                 ftrace_clear_events();
 276
 277         seq_ops = inode->i_private;
 278         ret = seq_open(file, seq_ops);
 279         if (!ret) {
 280                 struct seq_file *m = file->private_data;
 281
 282                 m->private = __start_ftrace_events;
 283         }
 284         return ret;
 285 }
 286
 287 static ssize_t
 288 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 289                   loff_t *ppos)
 290 {
 291         struct ftrace_event_call *call = filp->private_data;
 292         char *buf;
 293
 294         if (call->enabled)
 295                 buf = "1\n";
 296         else
 297                 buf = "0\n";
 298
 299         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 300 }
 301
 302 static ssize_t
 303 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 304                    loff_t *ppos)
 305 {
 306         struct ftrace_event_call *call = filp->private_data;
 307         char buf[64];
 308         unsigned long val;
 309         int ret;
 310
 311         if (cnt >= sizeof(buf))
 312                 return -EINVAL;
 313
 314         if (copy_from_user(&buf, ubuf, cnt))
 315                 return -EFAULT;
 316
 317         buf[cnt] = 0;
 318
 319         ret = strict_strtoul(buf, 10, &val);
 320         if (ret < 0)
 321                 return ret;
 322
 323         ret = tracing_update_buffers();
 324         if (ret < 0)
 325                 return ret;
 326
 327         switch (val) {
 328         case 0:
 329         case 1:
 330                 mutex_lock(&event_mutex);
 331                 ftrace_event_enable_disable(call, val);
 332                 mutex_unlock(&event_mutex);
 333                 break;
 334
 335         default:
 336                 return -EINVAL;
 337         }
 338
 339         *ppos += cnt;
 340
 341         return cnt;
 342 }
 343
 344 #undef FIELD
 345 #define FIELD(type, name)                                               \
 346         #type, #name, offsetof(typeof(field), name), sizeof(field.name)
 347
 348 static int trace_write_header(struct trace_seq *s)
 349 {
 350         struct trace_entry field;
 351
 352         /* struct trace_entry */
 353         return trace_seq_printf(s,
 354                                 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
 355                                 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
 356                                 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
 357                                 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
 358                                 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
 359                                 "\n",
 360                                 FIELD(unsigned char, type),
 361                                 FIELD(unsigned char, flags),
 362                                 FIELD(unsigned char, preempt_count),
 363                                 FIELD(int, pid),
 364                                 FIELD(int, tgid));
 365 }
 366
 367 static ssize_t
 368 event_format_read(struct file *filp, char __user *ubuf, size_t cnt,
 369                   loff_t *ppos)
 370 {
 371         struct ftrace_event_call *call = filp->private_data;
 372         struct trace_seq *s;
 373         char *buf;
 374         int r;
 375
 376         if (*ppos)
 377                 return 0;
 378
 379         s = kmalloc(sizeof(*s), GFP_KERNEL);
 380         if (!s)
 381                 return -ENOMEM;
 382
 383         trace_seq_init(s);
 384
 385         /* If any of the first writes fail, so will the show_format. */
 386
 387         trace_seq_printf(s, "name: %s\n", call->name);
 388         trace_seq_printf(s, "ID: %d\n", call->id);
 389         trace_seq_printf(s, "format:\n");
 390         trace_write_header(s);
 391
 392         r = call->show_format(s);
 393         if (!r) {
 394                 /*
 395                  * ug!  The format output is bigger than a PAGE!!
 396                  */
 397                 buf = "FORMAT TOO BIG\n";
 398                 r = simple_read_from_buffer(ubuf, cnt, ppos,
 399                                               buf, strlen(buf));
 400                 goto out;
 401         }
 402
 403         r = simple_read_from_buffer(ubuf, cnt, ppos,
 404                                     s->buffer, s->len);
 405  out:
 406         kfree(s);
 407         return r;
 408 }
 409
 410 static ssize_t
 411 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 412 {
 413         struct ftrace_event_call *call = filp->private_data;
 414         struct trace_seq *s;
 415         int r;
 416
 417         if (*ppos)
 418                 return 0;
 419
 420         s = kmalloc(sizeof(*s), GFP_KERNEL);
 421         if (!s)
 422                 return -ENOMEM;
 423
 424         trace_seq_init(s);
 425         trace_seq_printf(s, "%d\n", call->id);
 426
 427         r = simple_read_from_buffer(ubuf, cnt, ppos,
 428                                     s->buffer, s->len);
 429         kfree(s);
 430         return r;
 431 }
 432
 433 static const struct seq_operations show_event_seq_ops = {
 434         .start = t_start,
 435         .next = t_next,
 436         .show = t_show,
 437         .stop = t_stop,
 438 };
 439
 440 static const struct seq_operations show_set_event_seq_ops = {
 441         .start = s_start,
 442         .next = s_next,
 443         .show = t_show,
 444         .stop = t_stop,
 445 };
 446
 447 static const struct file_operations ftrace_avail_fops = {
 448         .open = ftrace_event_seq_open,
 449         .read = seq_read,
 450         .llseek = seq_lseek,
 451         .release = seq_release,
 452 };
 453
 454 static const struct file_operations ftrace_set_event_fops = {
 455         .open = ftrace_event_seq_open,
 456         .read = seq_read,
 457         .write = ftrace_event_write,
 458         .llseek = seq_lseek,
 459         .release = seq_release,
 460 };
 461
 462 static const struct file_operations ftrace_enable_fops = {
 463         .open = tracing_open_generic,
 464         .read = event_enable_read,
 465         .write = event_enable_write,
 466 };
 467
 468 static const struct file_operations ftrace_event_format_fops = {
 469         .open = tracing_open_generic,
 470         .read = event_format_read,
 471 };
 472
 473 static const struct file_operations ftrace_event_id_fops = {
 474         .open = tracing_open_generic,
 475         .read = event_id_read,
 476 };
 477
 478 static struct dentry *event_trace_events_dir(void)
 479 {
 480         static struct dentry *d_tracer;
 481         static struct dentry *d_events;
 482
 483         if (d_events)
 484                 return d_events;
 485
 486         d_tracer = tracing_init_dentry();
 487         if (!d_tracer)
 488                 return NULL;
 489
 490         d_events = debugfs_create_dir("events", d_tracer);
 491         if (!d_events)
 492                 pr_warning("Could not create debugfs "
 493                            "'events' directory\n");
 494
 495         return d_events;
 496 }
 497
 498 struct event_subsystem {
 499         struct list_head        list;
 500         const char              *name;
 501         struct dentry           *entry;
 502 };
 503
 504 static LIST_HEAD(event_subsystems);
 505
 506 static struct dentry *
 507 event_subsystem_dir(const char *name, struct dentry *d_events)
 508 {
 509         struct event_subsystem *system;
 510
 511         /* First see if we did not already create this dir */
 512         list_for_each_entry(system, &event_subsystems, list) {
 513                 if (strcmp(system->name, name) == 0)
 514                         return system->entry;
 515         }
 516
 517         /* need to create new entry */
 518         system = kmalloc(sizeof(*system), GFP_KERNEL);
 519         if (!system) {
 520                 pr_warning("No memory to create event subsystem %s\n",
 521                            name);
 522                 return d_events;
 523         }
 524
 525         system->entry = debugfs_create_dir(name, d_events);
 526         if (!system->entry) {
 527                 pr_warning("Could not create event subsystem %s\n",
 528                            name);
 529                 kfree(system);
 530                 return d_events;
 531         }
 532
 533         system->name = name;
 534         list_add(&system->list, &event_subsystems);
 535
 536         return system->entry;
 537 }
 538
 539 static int
 540 event_create_dir(struct ftrace_event_call *call, struct dentry *d_events)
 541 {
 542         struct dentry *entry;
 543         int ret;
 544
 545         /*
 546          * If the trace point header did not define TRACE_SYSTEM
 547          * then the system would be called "TRACE_SYSTEM".
 548          */
 549         if (strcmp(call->system, "TRACE_SYSTEM") != 0)
 550                 d_events = event_subsystem_dir(call->system, d_events);
 551
 552         if (call->raw_init) {
 553                 ret = call->raw_init();
 554                 if (ret < 0) {
 555                         pr_warning("Could not initialize trace point"
 556                                    " events/%s\n", call->name);
 557                         return ret;
 558                 }
 559         }
 560
 561         call->dir = debugfs_create_dir(call->name, d_events);
 562         if (!call->dir) {
 563                 pr_warning("Could not create debugfs "
 564                            "'%s' directory\n", call->name);
 565                 return -1;
 566         }
 567
 568         if (call->regfunc) {
 569                 entry = debugfs_create_file("enable", 0644, call->dir, call,
 570                                             &ftrace_enable_fops);
 571                 if (!entry)
 572                         pr_warning("Could not create debugfs "
 573                                    "'%s/enable' entry\n", call->name);
 574         }
 575
 576         if (call->id) {
 577                 entry = debugfs_create_file("id", 0444, call->dir, call,
 578                                 &ftrace_event_id_fops);
 579                 if (!entry)
 580                         pr_warning("Could not create debugfs '%s/id' entry\n",
 581                                         call->name);
 582         }
 583
 584         /* A trace may not want to export its format */
 585         if (!call->show_format)
 586                 return 0;
 587
 588         entry = debugfs_create_file("format", 0444, call->dir, call,
 589                                     &ftrace_event_format_fops);
 590         if (!entry)
 591                 pr_warning("Could not create debugfs "
 592                            "'%s/format' entry\n", call->name);
 593
 594         return 0;
 595 }
 596
 597 static __init int event_trace_init(void)
 598 {
 599         struct ftrace_event_call *call = __start_ftrace_events;
 600         struct dentry *d_tracer;
 601         struct dentry *entry;
 602         struct dentry *d_events;
 603
 604         d_tracer = tracing_init_dentry();
 605         if (!d_tracer)
 606                 return 0;
 607
 608         entry = debugfs_create_file("available_events", 0444, d_tracer,
 609                                     (void *)&show_event_seq_ops,
 610                                     &ftrace_avail_fops);
 611         if (!entry)
 612                 pr_warning("Could not create debugfs "
 613                            "'available_events' entry\n");
 614
 615         entry = debugfs_create_file("set_event", 0644, d_tracer,
 616                                     (void *)&show_set_event_seq_ops,
 617                                     &ftrace_set_event_fops);
 618         if (!entry)
 619                 pr_warning("Could not create debugfs "
 620                            "'set_event' entry\n");
 621
 622         d_events = event_trace_events_dir();
 623         if (!d_events)
 624                 return 0;
 625
 626         for_each_event(call) {
 627                 /* The linker may leave blanks */
 628                 if (!call->name)
 629                         continue;
 630                 event_create_dir(call, d_events);
 631         }
 632
 633         return 0;
 634 }
 635 fs_initcall(event_trace_init);