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