kernel/tracepoint.c

   1 /*
   2  * Copyright (C) 2008 Mathieu Desnoyers
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License as published by
   6  * the Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  17  */
  18 #include <linux/module.h>
  19 #include <linux/mutex.h>
  20 #include <linux/types.h>
  21 #include <linux/jhash.h>
  22 #include <linux/list.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/tracepoint.h>
  25 #include <linux/err.h>
  26 #include <linux/slab.h>
  27 #include <linux/sched.h>
  28 #include <linux/jump_label.h>
  29
  30 extern struct tracepoint * const __start___tracepoints_ptrs[];
  31 extern struct tracepoint * const __stop___tracepoints_ptrs[];
  32
  33 /* Set to 1 to enable tracepoint debug output */
  34 static const int tracepoint_debug;
  35
  36 /*
  37  * Tracepoints mutex protects the builtin and module tracepoints and the hash
  38  * table, as well as the local module list.
  39  */
  40 static DEFINE_MUTEX(tracepoints_mutex);
  41
  42 #ifdef CONFIG_MODULES
  43 /* Local list of struct module */
  44 static LIST_HEAD(tracepoint_module_list);
  45 #endif /* CONFIG_MODULES */
  46
  47 /*
  48  * Tracepoint hash table, containing the active tracepoints.
  49  * Protected by tracepoints_mutex.
  50  */
  51 #define TRACEPOINT_HASH_BITS 6
  52 #define TRACEPOINT_TABLE_SIZE (1 << TRACEPOINT_HASH_BITS)
  53 static struct hlist_head tracepoint_table[TRACEPOINT_TABLE_SIZE];
  54
  55 /*
  56  * Note about RCU :
  57  * It is used to delay the free of multiple probes array until a quiescent
  58  * state is reached.
  59  * Tracepoint entries modifications are protected by the tracepoints_mutex.
  60  */
  61 struct tracepoint_entry {
  62         struct hlist_node hlist;
  63         struct tracepoint_func *funcs;
  64         int refcount;   /* Number of times armed. 0 if disarmed. */
  65         char name[0];
  66 };
  67
  68 struct tp_probes {
  69         union {
  70                 struct rcu_head rcu;
  71                 struct list_head list;
  72         } u;
  73         struct tracepoint_func probes[0];
  74 };
  75
  76 static inline void *allocate_probes(int count)
  77 {
  78         struct tp_probes *p  = kmalloc(count * sizeof(struct tracepoint_func)
  79                         + sizeof(struct tp_probes), GFP_KERNEL);
  80         return p == NULL ? NULL : p->probes;
  81 }
  82
  83 static void rcu_free_old_probes(struct rcu_head *head)
  84 {
  85         kfree(container_of(head, struct tp_probes, u.rcu));
  86 }
  87
  88 static inline void release_probes(struct tracepoint_func *old)
  89 {
  90         if (old) {
  91                 struct tp_probes *tp_probes = container_of(old,
  92                         struct tp_probes, probes[0]);
  93                 call_rcu_sched(&tp_probes->u.rcu, rcu_free_old_probes);
  94         }
  95 }
  96
  97 static void debug_print_probes(struct tracepoint_entry *entry)
  98 {
  99         int i;
 100
 101         if (!tracepoint_debug || !entry->funcs)
 102                 return;
 103
 104         for (i = 0; entry->funcs[i].func; i++)
 105                 printk(KERN_DEBUG "Probe %d : %p\n", i, entry->funcs[i].func);
 106 }
 107
 108 static struct tracepoint_func *
 109 tracepoint_entry_add_probe(struct tracepoint_entry *entry,
 110                            void *probe, void *data)
 111 {
 112         int nr_probes = 0;
 113         struct tracepoint_func *old, *new;
 114
 115         WARN_ON(!probe);
 116
 117         debug_print_probes(entry);
 118         old = entry->funcs;
 119         if (old) {
 120                 /* (N -> N+1), (N != 0, 1) probes */
 121                 for (nr_probes = 0; old[nr_probes].func; nr_probes++)
 122                         if (old[nr_probes].func == probe &&
 123                             old[nr_probes].data == data)
 124                                 return ERR_PTR(-EEXIST);
 125         }
 126         /* + 2 : one for new probe, one for NULL func */
 127         new = allocate_probes(nr_probes + 2);
 128         if (new == NULL)
 129                 return ERR_PTR(-ENOMEM);
 130         if (old)
 131                 memcpy(new, old, nr_probes * sizeof(struct tracepoint_func));
 132         new[nr_probes].func = probe;
 133         new[nr_probes].data = data;
 134         new[nr_probes + 1].func = NULL;
 135         entry->refcount = nr_probes + 1;
 136         entry->funcs = new;
 137         debug_print_probes(entry);
 138         return old;
 139 }
 140
 141 static void *
 142 tracepoint_entry_remove_probe(struct tracepoint_entry *entry,
 143                               void *probe, void *data)
 144 {
 145         int nr_probes = 0, nr_del = 0, i;
 146         struct tracepoint_func *old, *new;
 147
 148         old = entry->funcs;
 149
 150         if (!old)
 151                 return ERR_PTR(-ENOENT);
 152
 153         debug_print_probes(entry);
 154         /* (N -> M), (N > 1, M >= 0) probes */
 155         for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
 156                 if (!probe ||
 157                     (old[nr_probes].func == probe &&
 158                      old[nr_probes].data == data))
 159                         nr_del++;
 160         }
 161
 162         if (nr_probes - nr_del == 0) {
 163                 /* N -> 0, (N > 1) */
 164                 entry->funcs = NULL;
 165                 entry->refcount = 0;
 166                 debug_print_probes(entry);
 167                 return old;
 168         } else {
 169                 int j = 0;
 170                 /* N -> M, (N > 1, M > 0) */
 171                 /* + 1 for NULL */
 172                 new = allocate_probes(nr_probes - nr_del + 1);
 173                 if (new == NULL)
 174                         return ERR_PTR(-ENOMEM);
 175                 for (i = 0; old[i].func; i++)
 176                         if (probe &&
 177                             (old[i].func != probe || old[i].data != data))
 178                                 new[j++] = old[i];
 179                 new[nr_probes - nr_del].func = NULL;
 180                 entry->refcount = nr_probes - nr_del;
 181                 entry->funcs = new;
 182         }
 183         debug_print_probes(entry);
 184         return old;
 185 }
 186
 187 /*
 188  * Get tracepoint if the tracepoint is present in the tracepoint hash table.
 189  * Must be called with tracepoints_mutex held.
 190  * Returns NULL if not present.
 191  */
 192 static struct tracepoint_entry *get_tracepoint(const char *name)
 193 {
 194         struct hlist_head *head;
 195         struct hlist_node *node;
 196         struct tracepoint_entry *e;
 197         u32 hash = jhash(name, strlen(name), 0);
 198
 199         head = &tracepoint_table[hash & (TRACEPOINT_TABLE_SIZE - 1)];
 200         hlist_for_each_entry(e, node, head, hlist) {
 201                 if (!strcmp(name, e->name))
 202                         return e;
 203         }
 204         return NULL;
 205 }
 206
 207 /*
 208  * Add the tracepoint to the tracepoint hash table. Must be called with
 209  * tracepoints_mutex held.
 210  */
 211 static struct tracepoint_entry *add_tracepoint(const char *name)
 212 {
 213         struct hlist_head *head;
 214         struct hlist_node *node;
 215         struct tracepoint_entry *e;
 216         size_t name_len = strlen(name) + 1;
 217         u32 hash = jhash(name, name_len-1, 0);
 218
 219         head = &tracepoint_table[hash & (TRACEPOINT_TABLE_SIZE - 1)];
 220         hlist_for_each_entry(e, node, head, hlist) {
 221                 if (!strcmp(name, e->name)) {
 222                         printk(KERN_NOTICE
 223                                 "tracepoint %s busy\n", name);
 224                         return ERR_PTR(-EEXIST);        /* Already there */
 225                 }
 226         }
 227         /*
 228          * Using kmalloc here to allocate a variable length element. Could
 229          * cause some memory fragmentation if overused.
 230          */
 231         e = kmalloc(sizeof(struct tracepoint_entry) + name_len, GFP_KERNEL);
 232         if (!e)
 233                 return ERR_PTR(-ENOMEM);
 234         memcpy(&e->name[0], name, name_len);
 235         e->funcs = NULL;
 236         e->refcount = 0;
 237         hlist_add_head(&e->hlist, head);
 238         return e;
 239 }
 240
 241 /*
 242  * Remove the tracepoint from the tracepoint hash table. Must be called with
 243  * mutex_lock held.
 244  */
 245 static inline void remove_tracepoint(struct tracepoint_entry *e)
 246 {
 247         hlist_del(&e->hlist);
 248         kfree(e);
 249 }
 250
 251 /*
 252  * Sets the probe callback corresponding to one tracepoint.
 253  */
 254 static void set_tracepoint(struct tracepoint_entry **entry,
 255         struct tracepoint *elem, int active)
 256 {
 257         WARN_ON(strcmp((*entry)->name, elem->name) != 0);
 258
 259         if (elem->regfunc && !jump_label_enabled(&elem->key) && active)
 260                 elem->regfunc();
 261         else if (elem->unregfunc && jump_label_enabled(&elem->key) && !active)
 262                 elem->unregfunc();
 263
 264         /*
 265          * rcu_assign_pointer has a smp_wmb() which makes sure that the new
 266          * probe callbacks array is consistent before setting a pointer to it.
 267          * This array is referenced by __DO_TRACE from
 268          * include/linux/tracepoints.h. A matching smp_read_barrier_depends()
 269          * is used.
 270          */
 271         rcu_assign_pointer(elem->funcs, (*entry)->funcs);
 272         if (active && !jump_label_enabled(&elem->key))
 273                 jump_label_inc(&elem->key);
 274         else if (!active && jump_label_enabled(&elem->key))
 275                 jump_label_dec(&elem->key);
 276 }
 277
 278 /*
 279  * Disable a tracepoint and its probe callback.
 280  * Note: only waiting an RCU period after setting elem->call to the empty
 281  * function insures that the original callback is not used anymore. This insured
 282  * by preempt_disable around the call site.
 283  */
 284 static void disable_tracepoint(struct tracepoint *elem)
 285 {
 286         if (elem->unregfunc && jump_label_enabled(&elem->key))
 287                 elem->unregfunc();
 288
 289         if (jump_label_enabled(&elem->key))
 290                 jump_label_dec(&elem->key);
 291         rcu_assign_pointer(elem->funcs, NULL);
 292 }
 293
 294 /**
 295  * tracepoint_update_probe_range - Update a probe range
 296  * @begin: beginning of the range
 297  * @end: end of the range
 298  *
 299  * Updates the probe callback corresponding to a range of tracepoints.
 300  * Called with tracepoints_mutex held.
 301  */
 302 static void tracepoint_update_probe_range(struct tracepoint * const *begin,
 303                                           struct tracepoint * const *end)
 304 {
 305         struct tracepoint * const *iter;
 306         struct tracepoint_entry *mark_entry;
 307
 308         if (!begin)
 309                 return;
 310
 311         for (iter = begin; iter < end; iter++) {
 312                 mark_entry = get_tracepoint((*iter)->name);
 313                 if (mark_entry) {
 314                         set_tracepoint(&mark_entry, *iter,
 315                                         !!mark_entry->refcount);
 316                 } else {
 317                         disable_tracepoint(*iter);
 318                 }
 319         }
 320 }
 321
 322 #ifdef CONFIG_MODULES
 323 void module_update_tracepoints(void)
 324 {
 325         struct tp_module *tp_mod;
 326
 327         list_for_each_entry(tp_mod, &tracepoint_module_list, list)
 328                 tracepoint_update_probe_range(tp_mod->tracepoints_ptrs,
 329                         tp_mod->tracepoints_ptrs + tp_mod->num_tracepoints);
 330 }
 331 #else /* CONFIG_MODULES */
 332 void module_update_tracepoints(void)
 333 {
 334 }
 335 #endif /* CONFIG_MODULES */
 336
 337
 338 /*
 339  * Update probes, removing the faulty probes.
 340  * Called with tracepoints_mutex held.
 341  */
 342 static void tracepoint_update_probes(void)
 343 {
 344         /* Core kernel tracepoints */
 345         tracepoint_update_probe_range(__start___tracepoints_ptrs,
 346                 __stop___tracepoints_ptrs);
 347         /* tracepoints in modules. */
 348         module_update_tracepoints();
 349 }
 350
 351 static struct tracepoint_func *
 352 tracepoint_add_probe(const char *name, void *probe, void *data)
 353 {
 354         struct tracepoint_entry *entry;
 355         struct tracepoint_func *old;
 356
 357         entry = get_tracepoint(name);
 358         if (!entry) {
 359                 entry = add_tracepoint(name);
 360                 if (IS_ERR(entry))
 361                         return (struct tracepoint_func *)entry;
 362         }
 363         old = tracepoint_entry_add_probe(entry, probe, data);
 364         if (IS_ERR(old) && !entry->refcount)
 365                 remove_tracepoint(entry);
 366         return old;
 367 }
 368
 369 /**
 370  * tracepoint_probe_register -  Connect a probe to a tracepoint
 371  * @name: tracepoint name
 372  * @probe: probe handler
 373  *
 374  * Returns 0 if ok, error value on error.
 375  * The probe address must at least be aligned on the architecture pointer size.
 376  */
 377 int tracepoint_probe_register(const char *name, void *probe, void *data)
 378 {
 379         struct tracepoint_func *old;
 380
 381         mutex_lock(&tracepoints_mutex);
 382         old = tracepoint_add_probe(name, probe, data);
 383         if (IS_ERR(old)) {
 384                 mutex_unlock(&tracepoints_mutex);
 385                 return PTR_ERR(old);
 386         }
 387         tracepoint_update_probes();             /* may update entry */
 388         mutex_unlock(&tracepoints_mutex);
 389         release_probes(old);
 390         return 0;
 391 }
 392 EXPORT_SYMBOL_GPL(tracepoint_probe_register);
 393
 394 static struct tracepoint_func *
 395 tracepoint_remove_probe(const char *name, void *probe, void *data)
 396 {
 397         struct tracepoint_entry *entry;
 398         struct tracepoint_func *old;
 399
 400         entry = get_tracepoint(name);
 401         if (!entry)
 402                 return ERR_PTR(-ENOENT);
 403         old = tracepoint_entry_remove_probe(entry, probe, data);
 404         if (IS_ERR(old))
 405                 return old;
 406         if (!entry->refcount)
 407                 remove_tracepoint(entry);
 408         return old;
 409 }
 410
 411 /**
 412  * tracepoint_probe_unregister -  Disconnect a probe from a tracepoint
 413  * @name: tracepoint name
 414  * @probe: probe function pointer
 415  *
 416  * We do not need to call a synchronize_sched to make sure the probes have
 417  * finished running before doing a module unload, because the module unload
 418  * itself uses stop_machine(), which insures that every preempt disabled section
 419  * have finished.
 420  */
 421 int tracepoint_probe_unregister(const char *name, void *probe, void *data)
 422 {
 423         struct tracepoint_func *old;
 424
 425         mutex_lock(&tracepoints_mutex);
 426         old = tracepoint_remove_probe(name, probe, data);
 427         if (IS_ERR(old)) {
 428                 mutex_unlock(&tracepoints_mutex);
 429                 return PTR_ERR(old);
 430         }
 431         tracepoint_update_probes();             /* may update entry */
 432         mutex_unlock(&tracepoints_mutex);
 433         release_probes(old);
 434         return 0;
 435 }
 436 EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
 437
 438 static LIST_HEAD(old_probes);
 439 static int need_update;
 440
 441 static void tracepoint_add_old_probes(void *old)
 442 {
 443         need_update = 1;
 444         if (old) {
 445                 struct tp_probes *tp_probes = container_of(old,
 446                         struct tp_probes, probes[0]);
 447                 list_add(&tp_probes->u.list, &old_probes);
 448         }
 449 }
 450
 451 /**
 452  * tracepoint_probe_register_noupdate -  register a probe but not connect
 453  * @name: tracepoint name
 454  * @probe: probe handler
 455  *
 456  * caller must call tracepoint_probe_update_all()
 457  */
 458 int tracepoint_probe_register_noupdate(const char *name, void *probe,
 459                                        void *data)
 460 {
 461         struct tracepoint_func *old;
 462
 463         mutex_lock(&tracepoints_mutex);
 464         old = tracepoint_add_probe(name, probe, data);
 465         if (IS_ERR(old)) {
 466                 mutex_unlock(&tracepoints_mutex);
 467                 return PTR_ERR(old);
 468         }
 469         tracepoint_add_old_probes(old);
 470         mutex_unlock(&tracepoints_mutex);
 471         return 0;
 472 }
 473 EXPORT_SYMBOL_GPL(tracepoint_probe_register_noupdate);
 474
 475 /**
 476  * tracepoint_probe_unregister_noupdate -  remove a probe but not disconnect
 477  * @name: tracepoint name
 478  * @probe: probe function pointer
 479  *
 480  * caller must call tracepoint_probe_update_all()
 481  */
 482 int tracepoint_probe_unregister_noupdate(const char *name, void *probe,
 483                                          void *data)
 484 {
 485         struct tracepoint_func *old;
 486
 487         mutex_lock(&tracepoints_mutex);
 488         old = tracepoint_remove_probe(name, probe, data);
 489         if (IS_ERR(old)) {
 490                 mutex_unlock(&tracepoints_mutex);
 491                 return PTR_ERR(old);
 492         }
 493         tracepoint_add_old_probes(old);
 494         mutex_unlock(&tracepoints_mutex);
 495         return 0;
 496 }
 497 EXPORT_SYMBOL_GPL(tracepoint_probe_unregister_noupdate);
 498
 499 /**
 500  * tracepoint_probe_update_all -  update tracepoints
 501  */
 502 void tracepoint_probe_update_all(void)
 503 {
 504         LIST_HEAD(release_probes);
 505         struct tp_probes *pos, *next;
 506
 507         mutex_lock(&tracepoints_mutex);
 508         if (!need_update) {
 509                 mutex_unlock(&tracepoints_mutex);
 510                 return;
 511         }
 512         if (!list_empty(&old_probes))
 513                 list_replace_init(&old_probes, &release_probes);
 514         need_update = 0;
 515         tracepoint_update_probes();
 516         mutex_unlock(&tracepoints_mutex);
 517         list_for_each_entry_safe(pos, next, &release_probes, u.list) {
 518                 list_del(&pos->u.list);
 519                 call_rcu_sched(&pos->u.rcu, rcu_free_old_probes);
 520         }
 521 }
 522 EXPORT_SYMBOL_GPL(tracepoint_probe_update_all);
 523
 524 /**
 525  * tracepoint_get_iter_range - Get a next tracepoint iterator given a range.
 526  * @tracepoint: current tracepoints (in), next tracepoint (out)
 527  * @begin: beginning of the range
 528  * @end: end of the range
 529  *
 530  * Returns whether a next tracepoint has been found (1) or not (0).
 531  * Will return the first tracepoint in the range if the input tracepoint is
 532  * NULL.
 533  */
 534 static int tracepoint_get_iter_range(struct tracepoint * const **tracepoint,
 535         struct tracepoint * const *begin, struct tracepoint * const *end)
 536 {
 537         if (!*tracepoint && begin != end) {
 538                 *tracepoint = begin;
 539                 return 1;
 540         }
 541         if (*tracepoint >= begin && *tracepoint < end)
 542                 return 1;
 543         return 0;
 544 }
 545
 546 #ifdef CONFIG_MODULES
 547 static void tracepoint_get_iter(struct tracepoint_iter *iter)
 548 {
 549         int found = 0;
 550         struct tp_module *iter_mod;
 551
 552         /* Core kernel tracepoints */
 553         if (!iter->module) {
 554                 found = tracepoint_get_iter_range(&iter->tracepoint,
 555                                 __start___tracepoints_ptrs,
 556                                 __stop___tracepoints_ptrs);
 557                 if (found)
 558                         goto end;
 559         }
 560         /* Tracepoints in modules */
 561         mutex_lock(&tracepoints_mutex);
 562         list_for_each_entry(iter_mod, &tracepoint_module_list, list) {
 563                 /*
 564                  * Sorted module list
 565                  */
 566                 if (iter_mod < iter->module)
 567                         continue;
 568                 else if (iter_mod > iter->module)
 569                         iter->tracepoint = NULL;
 570                 found = tracepoint_get_iter_range(&iter->tracepoint,
 571                         iter_mod->tracepoints_ptrs,
 572                         iter_mod->tracepoints_ptrs
 573                                 + iter_mod->num_tracepoints);
 574                 if (found) {
 575                         iter->module = iter_mod;
 576                         break;
 577                 }
 578         }
 579         mutex_unlock(&tracepoints_mutex);
 580 end:
 581         if (!found)
 582                 tracepoint_iter_reset(iter);
 583 }
 584 #else /* CONFIG_MODULES */
 585 static void tracepoint_get_iter(struct tracepoint_iter *iter)
 586 {
 587         int found = 0;
 588
 589         /* Core kernel tracepoints */
 590         found = tracepoint_get_iter_range(&iter->tracepoint,
 591                         __start___tracepoints_ptrs,
 592                         __stop___tracepoints_ptrs);
 593         if (!found)
 594                 tracepoint_iter_reset(iter);
 595 }
 596 #endif /* CONFIG_MODULES */
 597
 598 void tracepoint_iter_start(struct tracepoint_iter *iter)
 599 {
 600         tracepoint_get_iter(iter);
 601 }
 602 EXPORT_SYMBOL_GPL(tracepoint_iter_start);
 603
 604 void tracepoint_iter_next(struct tracepoint_iter *iter)
 605 {
 606         iter->tracepoint++;
 607         /*
 608          * iter->tracepoint may be invalid because we blindly incremented it.
 609          * Make sure it is valid by marshalling on the tracepoints, getting the
 610          * tracepoints from following modules if necessary.
 611          */
 612         tracepoint_get_iter(iter);
 613 }
 614 EXPORT_SYMBOL_GPL(tracepoint_iter_next);
 615
 616 void tracepoint_iter_stop(struct tracepoint_iter *iter)
 617 {
 618 }
 619 EXPORT_SYMBOL_GPL(tracepoint_iter_stop);
 620
 621 void tracepoint_iter_reset(struct tracepoint_iter *iter)
 622 {
 623 #ifdef CONFIG_MODULES
 624         iter->module = NULL;
 625 #endif /* CONFIG_MODULES */
 626         iter->tracepoint = NULL;
 627 }
 628 EXPORT_SYMBOL_GPL(tracepoint_iter_reset);
 629
 630 #ifdef CONFIG_MODULES
 631 bool trace_module_has_bad_taint(struct module *mod)
 632 {
 633         return mod->taints & ~((1 << TAINT_OOT_MODULE) | (1 << TAINT_CRAP));
 634 }
 635
 636 static int tracepoint_module_coming(struct module *mod)
 637 {
 638         struct tp_module *tp_mod, *iter;
 639         int ret = 0;
 640
 641         if (!mod->num_tracepoints)
 642                 return 0;
 643
 644         /*
 645          * We skip modules that taint the kernel, especially those with different
 646          * module headers (for forced load), to make sure we don't cause a crash.
 647          * Staging and out-of-tree GPL modules are fine.
 648          */
 649         if (trace_module_has_bad_taint(mod))
 650                 return 0;
 651         mutex_lock(&tracepoints_mutex);
 652         tp_mod = kmalloc(sizeof(struct tp_module), GFP_KERNEL);
 653         if (!tp_mod) {
 654                 ret = -ENOMEM;
 655                 goto end;
 656         }
 657         tp_mod->num_tracepoints = mod->num_tracepoints;
 658         tp_mod->tracepoints_ptrs = mod->tracepoints_ptrs;
 659
 660         /*
 661          * tracepoint_module_list is kept sorted by struct module pointer
 662          * address for iteration on tracepoints from a seq_file that can release
 663          * the mutex between calls.
 664          */
 665         list_for_each_entry_reverse(iter, &tracepoint_module_list, list) {
 666                 BUG_ON(iter == tp_mod); /* Should never be in the list twice */
 667                 if (iter < tp_mod) {
 668                         /* We belong to the location right after iter. */
 669                         list_add(&tp_mod->list, &iter->list);
 670                         goto module_added;
 671                 }
 672         }
 673         /* We belong to the beginning of the list */
 674         list_add(&tp_mod->list, &tracepoint_module_list);
 675 module_added:
 676         tracepoint_update_probe_range(mod->tracepoints_ptrs,
 677                 mod->tracepoints_ptrs + mod->num_tracepoints);
 678 end:
 679         mutex_unlock(&tracepoints_mutex);
 680         return ret;
 681 }
 682
 683 static int tracepoint_module_going(struct module *mod)
 684 {
 685         struct tp_module *pos;
 686
 687         if (!mod->num_tracepoints)
 688                 return 0;
 689
 690         mutex_lock(&tracepoints_mutex);
 691         tracepoint_update_probe_range(mod->tracepoints_ptrs,
 692                 mod->tracepoints_ptrs + mod->num_tracepoints);
 693         list_for_each_entry(pos, &tracepoint_module_list, list) {
 694                 if (pos->tracepoints_ptrs == mod->tracepoints_ptrs) {
 695                         list_del(&pos->list);
 696                         kfree(pos);
 697                         break;
 698                 }
 699         }
 700         /*
 701          * In the case of modules that were tainted at "coming", we'll simply
 702          * walk through the list without finding it. We cannot use the "tainted"
 703          * flag on "going", in case a module taints the kernel only after being
 704          * loaded.
 705          */
 706         mutex_unlock(&tracepoints_mutex);
 707         return 0;
 708 }
 709
 710 int tracepoint_module_notify(struct notifier_block *self,
 711                              unsigned long val, void *data)
 712 {
 713         struct module *mod = data;
 714         int ret = 0;
 715
 716         switch (val) {
 717         case MODULE_STATE_COMING:
 718                 ret = tracepoint_module_coming(mod);
 719                 break;
 720         case MODULE_STATE_LIVE:
 721                 break;
 722         case MODULE_STATE_GOING:
 723                 ret = tracepoint_module_going(mod);
 724                 break;
 725         }
 726         return ret;
 727 }
 728
 729 struct notifier_block tracepoint_module_nb = {
 730         .notifier_call = tracepoint_module_notify,
 731         .priority = 0,
 732 };
 733
 734 static int init_tracepoints(void)
 735 {
 736         return register_module_notifier(&tracepoint_module_nb);
 737 }
 738 __initcall(init_tracepoints);
 739 #endif /* CONFIG_MODULES */
 740
 741 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
 742
 743 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
 744 static int sys_tracepoint_refcount;
 745
 746 void syscall_regfunc(void)
 747 {
 748         unsigned long flags;
 749         struct task_struct *g, *t;
 750
 751         if (!sys_tracepoint_refcount) {
 752                 read_lock_irqsave(&tasklist_lock, flags);
 753                 do_each_thread(g, t) {
 754                         /* Skip kernel threads. */
 755                         if (t->mm)
 756                                 set_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
 757                 } while_each_thread(g, t);
 758                 read_unlock_irqrestore(&tasklist_lock, flags);
 759         }
 760         sys_tracepoint_refcount++;
 761 }
 762
 763 void syscall_unregfunc(void)
 764 {
 765         unsigned long flags;
 766         struct task_struct *g, *t;
 767
 768         sys_tracepoint_refcount--;
 769         if (!sys_tracepoint_refcount) {
 770                 read_lock_irqsave(&tasklist_lock, flags);
 771                 do_each_thread(g, t) {
 772                         clear_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
 773                 } while_each_thread(g, t);
 774                 read_unlock_irqrestore(&tasklist_lock, flags);
 775         }
 776 }
 777 #endif