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
  29 extern struct tracepoint __start___tracepoints[];
  30 extern struct tracepoint __stop___tracepoints[];
  31
  32 /* Set to 1 to enable tracepoint debug output */
  33 static const int tracepoint_debug;
  34
  35 /*
  36  * tracepoints_mutex nests inside module_mutex. Tracepoints mutex protects the
  37  * builtin and module tracepoints and the hash table.
  38  */
  39 static DEFINE_MUTEX(tracepoints_mutex);
  40
  41 /*
  42  * Tracepoint hash table, containing the active tracepoints.
  43  * Protected by tracepoints_mutex.
  44  */
  45 #define TRACEPOINT_HASH_BITS 6
  46 #define TRACEPOINT_TABLE_SIZE (1 << TRACEPOINT_HASH_BITS)
  47 static struct hlist_head tracepoint_table[TRACEPOINT_TABLE_SIZE];
  48
  49 /*
  50  * Note about RCU :
  51  * It is used to delay the free of multiple probes array until a quiescent
  52  * state is reached.
  53  * Tracepoint entries modifications are protected by the tracepoints_mutex.
  54  */
  55 struct tracepoint_entry {
  56         struct hlist_node hlist;
  57         struct tracepoint_func *funcs;
  58         int refcount;   /* Number of times armed. 0 if disarmed. */
  59         char name[0];
  60 };
  61
  62 struct tp_probes {
  63         union {
  64                 struct rcu_head rcu;
  65                 struct list_head list;
  66         } u;
  67         struct tracepoint_func probes[0];
  68 };
  69
  70 static inline void *allocate_probes(int count)
  71 {
  72         struct tp_probes *p  = kmalloc(count * sizeof(struct tracepoint_func)
  73                         + sizeof(struct tp_probes), GFP_KERNEL);
  74         return p == NULL ? NULL : p->probes;
  75 }
  76
  77 static void rcu_free_old_probes(struct rcu_head *head)
  78 {
  79         kfree(container_of(head, struct tp_probes, u.rcu));
  80 }
  81
  82 static inline void release_probes(struct tracepoint_func *old)
  83 {
  84         if (old) {
  85                 struct tp_probes *tp_probes = container_of(old,
  86                         struct tp_probes, probes[0]);
  87                 call_rcu_sched(&tp_probes->u.rcu, rcu_free_old_probes);
  88         }
  89 }
  90
  91 static void debug_print_probes(struct tracepoint_entry *entry)
  92 {
  93         int i;
  94
  95         if (!tracepoint_debug || !entry->funcs)
  96                 return;
  97
  98         for (i = 0; entry->funcs[i].func; i++)
  99                 printk(KERN_DEBUG "Probe %d : %p\n", i, entry->funcs[i].func);
 100 }
 101
 102 static struct tracepoint_func *
 103 tracepoint_entry_add_probe(struct tracepoint_entry *entry,
 104                            void *probe, void *data)
 105 {
 106         int nr_probes = 0;
 107         struct tracepoint_func *old, *new;
 108
 109         WARN_ON(!probe);
 110
 111         debug_print_probes(entry);
 112         old = entry->funcs;
 113         if (old) {
 114                 /* (N -> N+1), (N != 0, 1) probes */
 115                 for (nr_probes = 0; old[nr_probes].func; nr_probes++)
 116                         if (old[nr_probes].func == probe &&
 117                             old[nr_probes].data == data)
 118                                 return ERR_PTR(-EEXIST);
 119         }
 120         /* + 2 : one for new probe, one for NULL func */
 121         new = allocate_probes(nr_probes + 2);
 122         if (new == NULL)
 123                 return ERR_PTR(-ENOMEM);
 124         if (old)
 125                 memcpy(new, old, nr_probes * sizeof(struct tracepoint_func));
 126         new[nr_probes].func = probe;
 127         new[nr_probes].data = data;
 128         new[nr_probes + 1].func = NULL;
 129         entry->refcount = nr_probes + 1;
 130         entry->funcs = new;
 131         debug_print_probes(entry);
 132         return old;
 133 }
 134
 135 static void *
 136 tracepoint_entry_remove_probe(struct tracepoint_entry *entry,
 137                               void *probe, void *data)
 138 {
 139         int nr_probes = 0, nr_del = 0, i;
 140         struct tracepoint_func *old, *new;
 141
 142         old = entry->funcs;
 143
 144         if (!old)
 145                 return ERR_PTR(-ENOENT);
 146
 147         debug_print_probes(entry);
 148         /* (N -> M), (N > 1, M >= 0) probes */
 149         for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
 150                 if (!probe ||
 151                     (old[nr_probes].func == probe &&
 152                      old[nr_probes].data == data))
 153                         nr_del++;
 154         }
 155
 156         if (nr_probes - nr_del == 0) {
 157                 /* N -> 0, (N > 1) */
 158                 entry->funcs = NULL;
 159                 entry->refcount = 0;
 160                 debug_print_probes(entry);
 161                 return old;
 162         } else {
 163                 int j = 0;
 164                 /* N -> M, (N > 1, M > 0) */
 165                 /* + 1 for NULL */
 166                 new = allocate_probes(nr_probes - nr_del + 1);
 167                 if (new == NULL)
 168                         return ERR_PTR(-ENOMEM);
 169                 for (i = 0; old[i].func; i++)
 170                         if (probe &&
 171                             (old[i].func != probe || old[i].data != data))
 172                                 new[j++] = old[i];
 173                 new[nr_probes - nr_del].func = NULL;
 174                 entry->refcount = nr_probes - nr_del;
 175                 entry->funcs = new;
 176         }
 177         debug_print_probes(entry);
 178         return old;
 179 }
 180
 181 /*
 182  * Get tracepoint if the tracepoint is present in the tracepoint hash table.
 183  * Must be called with tracepoints_mutex held.
 184  * Returns NULL if not present.
 185  */
 186 static struct tracepoint_entry *get_tracepoint(const char *name)
 187 {
 188         struct hlist_head *head;
 189         struct hlist_node *node;
 190         struct tracepoint_entry *e;
 191         u32 hash = jhash(name, strlen(name), 0);
 192
 193         head = &tracepoint_table[hash & (TRACEPOINT_TABLE_SIZE - 1)];
 194         hlist_for_each_entry(e, node, head, hlist) {
 195                 if (!strcmp(name, e->name))
 196                         return e;
 197         }
 198         return NULL;
 199 }
 200
 201 /*
 202  * Add the tracepoint to the tracepoint hash table. Must be called with
 203  * tracepoints_mutex held.
 204  */
 205 static struct tracepoint_entry *add_tracepoint(const char *name)
 206 {
 207         struct hlist_head *head;
 208         struct hlist_node *node;
 209         struct tracepoint_entry *e;
 210         size_t name_len = strlen(name) + 1;
 211         u32 hash = jhash(name, name_len-1, 0);
 212
 213         head = &tracepoint_table[hash & (TRACEPOINT_TABLE_SIZE - 1)];
 214         hlist_for_each_entry(e, node, head, hlist) {
 215                 if (!strcmp(name, e->name)) {
 216                         printk(KERN_NOTICE
 217                                 "tracepoint %s busy\n", name);
 218                         return ERR_PTR(-EEXIST);        /* Already there */
 219                 }
 220         }
 221         /*
 222          * Using kmalloc here to allocate a variable length element. Could
 223          * cause some memory fragmentation if overused.
 224          */
 225         e = kmalloc(sizeof(struct tracepoint_entry) + name_len, GFP_KERNEL);
 226         if (!e)
 227                 return ERR_PTR(-ENOMEM);
 228         memcpy(&e->name[0], name, name_len);
 229         e->funcs = NULL;
 230         e->refcount = 0;
 231         hlist_add_head(&e->hlist, head);
 232         return e;
 233 }
 234
 235 /*
 236  * Remove the tracepoint from the tracepoint hash table. Must be called with
 237  * mutex_lock held.
 238  */
 239 static inline void remove_tracepoint(struct tracepoint_entry *e)
 240 {
 241         hlist_del(&e->hlist);
 242         kfree(e);
 243 }
 244
 245 /*
 246  * Sets the probe callback corresponding to one tracepoint.
 247  */
 248 static void set_tracepoint(struct tracepoint_entry **entry,
 249         struct tracepoint *elem, int active)
 250 {
 251         WARN_ON(strcmp((*entry)->name, elem->name) != 0);
 252
 253         if (elem->regfunc && !elem->state && active)
 254                 elem->regfunc();
 255         else if (elem->unregfunc && elem->state && !active)
 256                 elem->unregfunc();
 257
 258         /*
 259          * rcu_assign_pointer has a smp_wmb() which makes sure that the new
 260          * probe callbacks array is consistent before setting a pointer to it.
 261          * This array is referenced by __DO_TRACE from
 262          * include/linux/tracepoints.h. A matching smp_read_barrier_depends()
 263          * is used.
 264          */
 265         rcu_assign_pointer(elem->funcs, (*entry)->funcs);
 266         elem->state = active;
 267 }
 268
 269 /*
 270  * Disable a tracepoint and its probe callback.
 271  * Note: only waiting an RCU period after setting elem->call to the empty
 272  * function insures that the original callback is not used anymore. This insured
 273  * by preempt_disable around the call site.
 274  */
 275 static void disable_tracepoint(struct tracepoint *elem)
 276 {
 277         if (elem->unregfunc && elem->state)
 278                 elem->unregfunc();
 279
 280         elem->state = 0;
 281         rcu_assign_pointer(elem->funcs, NULL);
 282 }
 283
 284 /**
 285  * tracepoint_update_probe_range - Update a probe range
 286  * @begin: beginning of the range
 287  * @end: end of the range
 288  *
 289  * Updates the probe callback corresponding to a range of tracepoints.
 290  */
 291 void
 292 tracepoint_update_probe_range(struct tracepoint *begin, struct tracepoint *end)
 293 {
 294         struct tracepoint *iter;
 295         struct tracepoint_entry *mark_entry;
 296
 297         if (!begin)
 298                 return;
 299
 300         mutex_lock(&tracepoints_mutex);
 301         for (iter = begin; iter < end; iter++) {
 302                 mark_entry = get_tracepoint(iter->name);
 303                 if (mark_entry) {
 304                         set_tracepoint(&mark_entry, iter,
 305                                         !!mark_entry->refcount);
 306                 } else {
 307                         disable_tracepoint(iter);
 308                 }
 309         }
 310         mutex_unlock(&tracepoints_mutex);
 311 }
 312
 313 /*
 314  * Update probes, removing the faulty probes.
 315  */
 316 static void tracepoint_update_probes(void)
 317 {
 318         /* Core kernel tracepoints */
 319         tracepoint_update_probe_range(__start___tracepoints,
 320                 __stop___tracepoints);
 321         /* tracepoints in modules. */
 322         module_update_tracepoints();
 323 }
 324
 325 static struct tracepoint_func *
 326 tracepoint_add_probe(const char *name, void *probe, void *data)
 327 {
 328         struct tracepoint_entry *entry;
 329         struct tracepoint_func *old;
 330
 331         entry = get_tracepoint(name);
 332         if (!entry) {
 333                 entry = add_tracepoint(name);
 334                 if (IS_ERR(entry))
 335                         return (struct tracepoint_func *)entry;
 336         }
 337         old = tracepoint_entry_add_probe(entry, probe, data);
 338         if (IS_ERR(old) && !entry->refcount)
 339                 remove_tracepoint(entry);
 340         return old;
 341 }
 342
 343 /**
 344  * tracepoint_probe_register -  Connect a probe to a tracepoint
 345  * @name: tracepoint name
 346  * @probe: probe handler
 347  *
 348  * Returns 0 if ok, error value on error.
 349  * The probe address must at least be aligned on the architecture pointer size.
 350  */
 351 int tracepoint_probe_register(const char *name, void *probe, void *data)
 352 {
 353         struct tracepoint_func *old;
 354
 355         mutex_lock(&tracepoints_mutex);
 356         old = tracepoint_add_probe(name, probe, data);
 357         mutex_unlock(&tracepoints_mutex);
 358         if (IS_ERR(old))
 359                 return PTR_ERR(old);
 360
 361         tracepoint_update_probes();             /* may update entry */
 362         release_probes(old);
 363         return 0;
 364 }
 365 EXPORT_SYMBOL_GPL(tracepoint_probe_register);
 366
 367 static struct tracepoint_func *
 368 tracepoint_remove_probe(const char *name, void *probe, void *data)
 369 {
 370         struct tracepoint_entry *entry;
 371         struct tracepoint_func *old;
 372
 373         entry = get_tracepoint(name);
 374         if (!entry)
 375                 return ERR_PTR(-ENOENT);
 376         old = tracepoint_entry_remove_probe(entry, probe, data);
 377         if (IS_ERR(old))
 378                 return old;
 379         if (!entry->refcount)
 380                 remove_tracepoint(entry);
 381         return old;
 382 }
 383
 384 /**
 385  * tracepoint_probe_unregister -  Disconnect a probe from a tracepoint
 386  * @name: tracepoint name
 387  * @probe: probe function pointer
 388  *
 389  * We do not need to call a synchronize_sched to make sure the probes have
 390  * finished running before doing a module unload, because the module unload
 391  * itself uses stop_machine(), which insures that every preempt disabled section
 392  * have finished.
 393  */
 394 int tracepoint_probe_unregister(const char *name, void *probe, void *data)
 395 {
 396         struct tracepoint_func *old;
 397
 398         mutex_lock(&tracepoints_mutex);
 399         old = tracepoint_remove_probe(name, probe, data);
 400         mutex_unlock(&tracepoints_mutex);
 401         if (IS_ERR(old))
 402                 return PTR_ERR(old);
 403
 404         tracepoint_update_probes();             /* may update entry */
 405         release_probes(old);
 406         return 0;
 407 }
 408 EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
 409
 410 static LIST_HEAD(old_probes);
 411 static int need_update;
 412
 413 static void tracepoint_add_old_probes(void *old)
 414 {
 415         need_update = 1;
 416         if (old) {
 417                 struct tp_probes *tp_probes = container_of(old,
 418                         struct tp_probes, probes[0]);
 419                 list_add(&tp_probes->u.list, &old_probes);
 420         }
 421 }
 422
 423 /**
 424  * tracepoint_probe_register_noupdate -  register a probe but not connect
 425  * @name: tracepoint name
 426  * @probe: probe handler
 427  *
 428  * caller must call tracepoint_probe_update_all()
 429  */
 430 int tracepoint_probe_register_noupdate(const char *name, void *probe,
 431                                        void *data)
 432 {
 433         struct tracepoint_func *old;
 434
 435         mutex_lock(&tracepoints_mutex);
 436         old = tracepoint_add_probe(name, probe, data);
 437         if (IS_ERR(old)) {
 438                 mutex_unlock(&tracepoints_mutex);
 439                 return PTR_ERR(old);
 440         }
 441         tracepoint_add_old_probes(old);
 442         mutex_unlock(&tracepoints_mutex);
 443         return 0;
 444 }
 445 EXPORT_SYMBOL_GPL(tracepoint_probe_register_noupdate);
 446
 447 /**
 448  * tracepoint_probe_unregister_noupdate -  remove a probe but not disconnect
 449  * @name: tracepoint name
 450  * @probe: probe function pointer
 451  *
 452  * caller must call tracepoint_probe_update_all()
 453  */
 454 int tracepoint_probe_unregister_noupdate(const char *name, void *probe,
 455                                          void *data)
 456 {
 457         struct tracepoint_func *old;
 458
 459         mutex_lock(&tracepoints_mutex);
 460         old = tracepoint_remove_probe(name, probe, data);
 461         if (IS_ERR(old)) {
 462                 mutex_unlock(&tracepoints_mutex);
 463                 return PTR_ERR(old);
 464         }
 465         tracepoint_add_old_probes(old);
 466         mutex_unlock(&tracepoints_mutex);
 467         return 0;
 468 }
 469 EXPORT_SYMBOL_GPL(tracepoint_probe_unregister_noupdate);
 470
 471 /**
 472  * tracepoint_probe_update_all -  update tracepoints
 473  */
 474 void tracepoint_probe_update_all(void)
 475 {
 476         LIST_HEAD(release_probes);
 477         struct tp_probes *pos, *next;
 478
 479         mutex_lock(&tracepoints_mutex);
 480         if (!need_update) {
 481                 mutex_unlock(&tracepoints_mutex);
 482                 return;
 483         }
 484         if (!list_empty(&old_probes))
 485                 list_replace_init(&old_probes, &release_probes);
 486         need_update = 0;
 487         mutex_unlock(&tracepoints_mutex);
 488
 489         tracepoint_update_probes();
 490         list_for_each_entry_safe(pos, next, &release_probes, u.list) {
 491                 list_del(&pos->u.list);
 492                 call_rcu_sched(&pos->u.rcu, rcu_free_old_probes);
 493         }
 494 }
 495 EXPORT_SYMBOL_GPL(tracepoint_probe_update_all);
 496
 497 /**
 498  * tracepoint_get_iter_range - Get a next tracepoint iterator given a range.
 499  * @tracepoint: current tracepoints (in), next tracepoint (out)
 500  * @begin: beginning of the range
 501  * @end: end of the range
 502  *
 503  * Returns whether a next tracepoint has been found (1) or not (0).
 504  * Will return the first tracepoint in the range if the input tracepoint is
 505  * NULL.
 506  */
 507 int tracepoint_get_iter_range(struct tracepoint **tracepoint,
 508         struct tracepoint *begin, struct tracepoint *end)
 509 {
 510         if (!*tracepoint && begin != end) {
 511                 *tracepoint = begin;
 512                 return 1;
 513         }
 514         if (*tracepoint >= begin && *tracepoint < end)
 515                 return 1;
 516         return 0;
 517 }
 518 EXPORT_SYMBOL_GPL(tracepoint_get_iter_range);
 519
 520 static void tracepoint_get_iter(struct tracepoint_iter *iter)
 521 {
 522         int found = 0;
 523
 524         /* Core kernel tracepoints */
 525         if (!iter->module) {
 526                 found = tracepoint_get_iter_range(&iter->tracepoint,
 527                                 __start___tracepoints, __stop___tracepoints);
 528                 if (found)
 529                         goto end;
 530         }
 531         /* tracepoints in modules. */
 532         found = module_get_iter_tracepoints(iter);
 533 end:
 534         if (!found)
 535                 tracepoint_iter_reset(iter);
 536 }
 537
 538 void tracepoint_iter_start(struct tracepoint_iter *iter)
 539 {
 540         tracepoint_get_iter(iter);
 541 }
 542 EXPORT_SYMBOL_GPL(tracepoint_iter_start);
 543
 544 void tracepoint_iter_next(struct tracepoint_iter *iter)
 545 {
 546         iter->tracepoint++;
 547         /*
 548          * iter->tracepoint may be invalid because we blindly incremented it.
 549          * Make sure it is valid by marshalling on the tracepoints, getting the
 550          * tracepoints from following modules if necessary.
 551          */
 552         tracepoint_get_iter(iter);
 553 }
 554 EXPORT_SYMBOL_GPL(tracepoint_iter_next);
 555
 556 void tracepoint_iter_stop(struct tracepoint_iter *iter)
 557 {
 558 }
 559 EXPORT_SYMBOL_GPL(tracepoint_iter_stop);
 560
 561 void tracepoint_iter_reset(struct tracepoint_iter *iter)
 562 {
 563         iter->module = NULL;
 564         iter->tracepoint = NULL;
 565 }
 566 EXPORT_SYMBOL_GPL(tracepoint_iter_reset);
 567
 568 #ifdef CONFIG_MODULES
 569
 570 int tracepoint_module_notify(struct notifier_block *self,
 571                              unsigned long val, void *data)
 572 {
 573         struct module *mod = data;
 574
 575         switch (val) {
 576         case MODULE_STATE_COMING:
 577         case MODULE_STATE_GOING:
 578                 tracepoint_update_probe_range(mod->tracepoints,
 579                         mod->tracepoints + mod->num_tracepoints);
 580                 break;
 581         }
 582         return 0;
 583 }
 584
 585 struct notifier_block tracepoint_module_nb = {
 586         .notifier_call = tracepoint_module_notify,
 587         .priority = 0,
 588 };
 589
 590 static int init_tracepoints(void)
 591 {
 592         return register_module_notifier(&tracepoint_module_nb);
 593 }
 594 __initcall(init_tracepoints);
 595
 596 #endif /* CONFIG_MODULES */
 597
 598 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
 599
 600 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
 601 static int sys_tracepoint_refcount;
 602
 603 void syscall_regfunc(void)
 604 {
 605         unsigned long flags;
 606         struct task_struct *g, *t;
 607
 608         if (!sys_tracepoint_refcount) {
 609                 read_lock_irqsave(&tasklist_lock, flags);
 610                 do_each_thread(g, t) {
 611                         /* Skip kernel threads. */
 612                         if (t->mm)
 613                                 set_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
 614                 } while_each_thread(g, t);
 615                 read_unlock_irqrestore(&tasklist_lock, flags);
 616         }
 617         sys_tracepoint_refcount++;
 618 }
 619
 620 void syscall_unregfunc(void)
 621 {
 622         unsigned long flags;
 623         struct task_struct *g, *t;
 624
 625         sys_tracepoint_refcount--;
 626         if (!sys_tracepoint_refcount) {
 627                 read_lock_irqsave(&tasklist_lock, flags);
 628                 do_each_thread(g, t) {
 629                         clear_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
 630                 } while_each_thread(g, t);
 631                 read_unlock_irqrestore(&tasklist_lock, flags);
 632         }
 633 }
 634 #endif