2 * Copyright (C) 2008 Mathieu Desnoyers
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.
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.
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.
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/static_key.h>
30 extern struct tracepoint * const __start___tracepoints_ptrs[];
31 extern struct tracepoint * const __stop___tracepoints_ptrs[];
33 /* Set to 1 to enable tracepoint debug output */
34 static const int tracepoint_debug;
37 * Tracepoints mutex protects the builtin and module tracepoints and the hash
38 * table, as well as the local module list.
40 static DEFINE_MUTEX(tracepoints_mutex);
43 /* Local list of struct module */
44 static LIST_HEAD(tracepoint_module_list);
45 #endif /* CONFIG_MODULES */
48 * Tracepoint hash table, containing the active tracepoints.
49 * Protected by tracepoints_mutex.
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];
57 * It is used to delay the free of multiple probes array until a quiescent
59 * Tracepoint entries modifications are protected by the tracepoints_mutex.
61 struct tracepoint_entry {
62 struct hlist_node hlist;
63 struct tracepoint_func *funcs;
64 int refcount; /* Number of times armed. 0 if disarmed. */
71 struct list_head list;
73 struct tracepoint_func probes[0];
76 static inline void *allocate_probes(int count)
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;
83 static void rcu_free_old_probes(struct rcu_head *head)
85 kfree(container_of(head, struct tp_probes, u.rcu));
88 static inline void release_probes(struct tracepoint_func *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);
97 static void debug_print_probes(struct tracepoint_entry *entry)
101 if (!tracepoint_debug || !entry->funcs)
104 for (i = 0; entry->funcs[i].func; i++)
105 printk(KERN_DEBUG "Probe %d : %p\n", i, entry->funcs[i].func);
108 static struct tracepoint_func *
109 tracepoint_entry_add_probe(struct tracepoint_entry *entry,
110 void *probe, void *data)
113 struct tracepoint_func *old, *new;
116 return ERR_PTR(-EINVAL);
118 debug_print_probes(entry);
121 /* (N -> N+1), (N != 0, 1) probes */
122 for (nr_probes = 0; old[nr_probes].func; nr_probes++)
123 if (old[nr_probes].func == probe &&
124 old[nr_probes].data == data)
125 return ERR_PTR(-EEXIST);
127 /* + 2 : one for new probe, one for NULL func */
128 new = allocate_probes(nr_probes + 2);
130 return ERR_PTR(-ENOMEM);
132 memcpy(new, old, nr_probes * sizeof(struct tracepoint_func));
133 new[nr_probes].func = probe;
134 new[nr_probes].data = data;
135 new[nr_probes + 1].func = NULL;
136 entry->refcount = nr_probes + 1;
138 debug_print_probes(entry);
143 tracepoint_entry_remove_probe(struct tracepoint_entry *entry,
144 void *probe, void *data)
146 int nr_probes = 0, nr_del = 0, i;
147 struct tracepoint_func *old, *new;
152 return ERR_PTR(-ENOENT);
154 debug_print_probes(entry);
155 /* (N -> M), (N > 1, M >= 0) probes */
157 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
158 if (old[nr_probes].func == probe &&
159 old[nr_probes].data == data)
165 * If probe is NULL, then nr_probes = nr_del = 0, and then the
166 * entire entry will be removed.
168 if (nr_probes - nr_del == 0) {
169 /* N -> 0, (N > 1) */
172 debug_print_probes(entry);
176 /* N -> M, (N > 1, M > 0) */
178 new = allocate_probes(nr_probes - nr_del + 1);
180 return ERR_PTR(-ENOMEM);
181 for (i = 0; old[i].func; i++)
182 if (old[i].func != probe || old[i].data != data)
184 new[nr_probes - nr_del].func = NULL;
185 entry->refcount = nr_probes - nr_del;
188 debug_print_probes(entry);
193 * Get tracepoint if the tracepoint is present in the tracepoint hash table.
194 * Must be called with tracepoints_mutex held.
195 * Returns NULL if not present.
197 static struct tracepoint_entry *get_tracepoint(const char *name)
199 struct hlist_head *head;
200 struct tracepoint_entry *e;
201 u32 hash = jhash(name, strlen(name), 0);
203 head = &tracepoint_table[hash & (TRACEPOINT_TABLE_SIZE - 1)];
204 hlist_for_each_entry(e, head, hlist) {
205 if (!strcmp(name, e->name))
212 * Add the tracepoint to the tracepoint hash table. Must be called with
213 * tracepoints_mutex held.
215 static struct tracepoint_entry *add_tracepoint(const char *name)
217 struct hlist_head *head;
218 struct tracepoint_entry *e;
219 size_t name_len = strlen(name) + 1;
220 u32 hash = jhash(name, name_len-1, 0);
222 head = &tracepoint_table[hash & (TRACEPOINT_TABLE_SIZE - 1)];
223 hlist_for_each_entry(e, head, hlist) {
224 if (!strcmp(name, e->name)) {
226 "tracepoint %s busy\n", name);
227 return ERR_PTR(-EEXIST); /* Already there */
231 * Using kmalloc here to allocate a variable length element. Could
232 * cause some memory fragmentation if overused.
234 e = kmalloc(sizeof(struct tracepoint_entry) + name_len, GFP_KERNEL);
236 return ERR_PTR(-ENOMEM);
237 memcpy(&e->name[0], name, name_len);
240 hlist_add_head(&e->hlist, head);
245 * Remove the tracepoint from the tracepoint hash table. Must be called with
248 static inline void remove_tracepoint(struct tracepoint_entry *e)
250 hlist_del(&e->hlist);
255 * Sets the probe callback corresponding to one tracepoint.
257 static void set_tracepoint(struct tracepoint_entry **entry,
258 struct tracepoint *elem, int active)
260 WARN_ON(strcmp((*entry)->name, elem->name) != 0);
262 if (elem->regfunc && !static_key_enabled(&elem->key) && active)
264 else if (elem->unregfunc && static_key_enabled(&elem->key) && !active)
268 * rcu_assign_pointer has a smp_wmb() which makes sure that the new
269 * probe callbacks array is consistent before setting a pointer to it.
270 * This array is referenced by __DO_TRACE from
271 * include/linux/tracepoints.h. A matching smp_read_barrier_depends()
274 rcu_assign_pointer(elem->funcs, (*entry)->funcs);
275 if (active && !static_key_enabled(&elem->key))
276 static_key_slow_inc(&elem->key);
277 else if (!active && static_key_enabled(&elem->key))
278 static_key_slow_dec(&elem->key);
282 * Disable a tracepoint and its probe callback.
283 * Note: only waiting an RCU period after setting elem->call to the empty
284 * function insures that the original callback is not used anymore. This insured
285 * by preempt_disable around the call site.
287 static void disable_tracepoint(struct tracepoint *elem)
289 if (elem->unregfunc && static_key_enabled(&elem->key))
292 if (static_key_enabled(&elem->key))
293 static_key_slow_dec(&elem->key);
294 rcu_assign_pointer(elem->funcs, NULL);
298 * tracepoint_update_probe_range - Update a probe range
299 * @begin: beginning of the range
300 * @end: end of the range
302 * Updates the probe callback corresponding to a range of tracepoints.
303 * Called with tracepoints_mutex held.
305 static void tracepoint_update_probe_range(struct tracepoint * const *begin,
306 struct tracepoint * const *end)
308 struct tracepoint * const *iter;
309 struct tracepoint_entry *mark_entry;
314 for (iter = begin; iter < end; iter++) {
315 mark_entry = get_tracepoint((*iter)->name);
317 set_tracepoint(&mark_entry, *iter,
318 !!mark_entry->refcount);
320 disable_tracepoint(*iter);
325 #ifdef CONFIG_MODULES
326 void module_update_tracepoints(void)
328 struct tp_module *tp_mod;
330 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
331 tracepoint_update_probe_range(tp_mod->tracepoints_ptrs,
332 tp_mod->tracepoints_ptrs + tp_mod->num_tracepoints);
334 #else /* CONFIG_MODULES */
335 void module_update_tracepoints(void)
338 #endif /* CONFIG_MODULES */
342 * Update probes, removing the faulty probes.
343 * Called with tracepoints_mutex held.
345 static void tracepoint_update_probes(void)
347 /* Core kernel tracepoints */
348 tracepoint_update_probe_range(__start___tracepoints_ptrs,
349 __stop___tracepoints_ptrs);
350 /* tracepoints in modules. */
351 module_update_tracepoints();
354 static struct tracepoint_func *
355 tracepoint_add_probe(const char *name, void *probe, void *data)
357 struct tracepoint_entry *entry;
358 struct tracepoint_func *old;
360 entry = get_tracepoint(name);
362 entry = add_tracepoint(name);
364 return (struct tracepoint_func *)entry;
366 old = tracepoint_entry_add_probe(entry, probe, data);
367 if (IS_ERR(old) && !entry->refcount)
368 remove_tracepoint(entry);
373 * tracepoint_probe_register - Connect a probe to a tracepoint
374 * @name: tracepoint name
375 * @probe: probe handler
377 * Returns 0 if ok, error value on error.
378 * The probe address must at least be aligned on the architecture pointer size.
380 int tracepoint_probe_register(const char *name, void *probe, void *data)
382 struct tracepoint_func *old;
384 mutex_lock(&tracepoints_mutex);
385 old = tracepoint_add_probe(name, probe, data);
387 mutex_unlock(&tracepoints_mutex);
390 tracepoint_update_probes(); /* may update entry */
391 mutex_unlock(&tracepoints_mutex);
395 EXPORT_SYMBOL_GPL(tracepoint_probe_register);
397 static struct tracepoint_func *
398 tracepoint_remove_probe(const char *name, void *probe, void *data)
400 struct tracepoint_entry *entry;
401 struct tracepoint_func *old;
403 entry = get_tracepoint(name);
405 return ERR_PTR(-ENOENT);
406 old = tracepoint_entry_remove_probe(entry, probe, data);
409 if (!entry->refcount)
410 remove_tracepoint(entry);
415 * tracepoint_probe_unregister - Disconnect a probe from a tracepoint
416 * @name: tracepoint name
417 * @probe: probe function pointer
419 * We do not need to call a synchronize_sched to make sure the probes have
420 * finished running before doing a module unload, because the module unload
421 * itself uses stop_machine(), which insures that every preempt disabled section
424 int tracepoint_probe_unregister(const char *name, void *probe, void *data)
426 struct tracepoint_func *old;
428 mutex_lock(&tracepoints_mutex);
429 old = tracepoint_remove_probe(name, probe, data);
431 mutex_unlock(&tracepoints_mutex);
434 tracepoint_update_probes(); /* may update entry */
435 mutex_unlock(&tracepoints_mutex);
439 EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
441 static LIST_HEAD(old_probes);
442 static int need_update;
444 static void tracepoint_add_old_probes(void *old)
448 struct tp_probes *tp_probes = container_of(old,
449 struct tp_probes, probes[0]);
450 list_add(&tp_probes->u.list, &old_probes);
455 * tracepoint_probe_register_noupdate - register a probe but not connect
456 * @name: tracepoint name
457 * @probe: probe handler
459 * caller must call tracepoint_probe_update_all()
461 int tracepoint_probe_register_noupdate(const char *name, void *probe,
464 struct tracepoint_func *old;
466 mutex_lock(&tracepoints_mutex);
467 old = tracepoint_add_probe(name, probe, data);
469 mutex_unlock(&tracepoints_mutex);
472 tracepoint_add_old_probes(old);
473 mutex_unlock(&tracepoints_mutex);
476 EXPORT_SYMBOL_GPL(tracepoint_probe_register_noupdate);
479 * tracepoint_probe_unregister_noupdate - remove a probe but not disconnect
480 * @name: tracepoint name
481 * @probe: probe function pointer
483 * caller must call tracepoint_probe_update_all()
485 int tracepoint_probe_unregister_noupdate(const char *name, void *probe,
488 struct tracepoint_func *old;
490 mutex_lock(&tracepoints_mutex);
491 old = tracepoint_remove_probe(name, probe, data);
493 mutex_unlock(&tracepoints_mutex);
496 tracepoint_add_old_probes(old);
497 mutex_unlock(&tracepoints_mutex);
500 EXPORT_SYMBOL_GPL(tracepoint_probe_unregister_noupdate);
503 * tracepoint_probe_update_all - update tracepoints
505 void tracepoint_probe_update_all(void)
507 LIST_HEAD(release_probes);
508 struct tp_probes *pos, *next;
510 mutex_lock(&tracepoints_mutex);
512 mutex_unlock(&tracepoints_mutex);
515 if (!list_empty(&old_probes))
516 list_replace_init(&old_probes, &release_probes);
518 tracepoint_update_probes();
519 mutex_unlock(&tracepoints_mutex);
520 list_for_each_entry_safe(pos, next, &release_probes, u.list) {
521 list_del(&pos->u.list);
522 call_rcu_sched(&pos->u.rcu, rcu_free_old_probes);
525 EXPORT_SYMBOL_GPL(tracepoint_probe_update_all);
528 * tracepoint_get_iter_range - Get a next tracepoint iterator given a range.
529 * @tracepoint: current tracepoints (in), next tracepoint (out)
530 * @begin: beginning of the range
531 * @end: end of the range
533 * Returns whether a next tracepoint has been found (1) or not (0).
534 * Will return the first tracepoint in the range if the input tracepoint is
537 static int tracepoint_get_iter_range(struct tracepoint * const **tracepoint,
538 struct tracepoint * const *begin, struct tracepoint * const *end)
540 if (!*tracepoint && begin != end) {
544 if (*tracepoint >= begin && *tracepoint < end)
549 #ifdef CONFIG_MODULES
550 static void tracepoint_get_iter(struct tracepoint_iter *iter)
553 struct tp_module *iter_mod;
555 /* Core kernel tracepoints */
557 found = tracepoint_get_iter_range(&iter->tracepoint,
558 __start___tracepoints_ptrs,
559 __stop___tracepoints_ptrs);
563 /* Tracepoints in modules */
564 mutex_lock(&tracepoints_mutex);
565 list_for_each_entry(iter_mod, &tracepoint_module_list, list) {
569 if (iter_mod < iter->module)
571 else if (iter_mod > iter->module)
572 iter->tracepoint = NULL;
573 found = tracepoint_get_iter_range(&iter->tracepoint,
574 iter_mod->tracepoints_ptrs,
575 iter_mod->tracepoints_ptrs
576 + iter_mod->num_tracepoints);
578 iter->module = iter_mod;
582 mutex_unlock(&tracepoints_mutex);
585 tracepoint_iter_reset(iter);
587 #else /* CONFIG_MODULES */
588 static void tracepoint_get_iter(struct tracepoint_iter *iter)
592 /* Core kernel tracepoints */
593 found = tracepoint_get_iter_range(&iter->tracepoint,
594 __start___tracepoints_ptrs,
595 __stop___tracepoints_ptrs);
597 tracepoint_iter_reset(iter);
599 #endif /* CONFIG_MODULES */
601 void tracepoint_iter_start(struct tracepoint_iter *iter)
603 tracepoint_get_iter(iter);
605 EXPORT_SYMBOL_GPL(tracepoint_iter_start);
607 void tracepoint_iter_next(struct tracepoint_iter *iter)
611 * iter->tracepoint may be invalid because we blindly incremented it.
612 * Make sure it is valid by marshalling on the tracepoints, getting the
613 * tracepoints from following modules if necessary.
615 tracepoint_get_iter(iter);
617 EXPORT_SYMBOL_GPL(tracepoint_iter_next);
619 void tracepoint_iter_stop(struct tracepoint_iter *iter)
622 EXPORT_SYMBOL_GPL(tracepoint_iter_stop);
624 void tracepoint_iter_reset(struct tracepoint_iter *iter)
626 #ifdef CONFIG_MODULES
628 #endif /* CONFIG_MODULES */
629 iter->tracepoint = NULL;
631 EXPORT_SYMBOL_GPL(tracepoint_iter_reset);
633 #ifdef CONFIG_MODULES
634 bool trace_module_has_bad_taint(struct module *mod)
636 return mod->taints & ~((1 << TAINT_OOT_MODULE) | (1 << TAINT_CRAP));
639 static int tracepoint_module_coming(struct module *mod)
641 struct tp_module *tp_mod, *iter;
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.
649 if (trace_module_has_bad_taint(mod))
651 mutex_lock(&tracepoints_mutex);
652 tp_mod = kmalloc(sizeof(struct tp_module), GFP_KERNEL);
657 tp_mod->num_tracepoints = mod->num_tracepoints;
658 tp_mod->tracepoints_ptrs = mod->tracepoints_ptrs;
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.
665 list_for_each_entry_reverse(iter, &tracepoint_module_list, list) {
666 BUG_ON(iter == tp_mod); /* Should never be in the list twice */
668 /* We belong to the location right after iter. */
669 list_add(&tp_mod->list, &iter->list);
673 /* We belong to the beginning of the list */
674 list_add(&tp_mod->list, &tracepoint_module_list);
676 tracepoint_update_probe_range(mod->tracepoints_ptrs,
677 mod->tracepoints_ptrs + mod->num_tracepoints);
679 mutex_unlock(&tracepoints_mutex);
683 static int tracepoint_module_going(struct module *mod)
685 struct tp_module *pos;
687 mutex_lock(&tracepoints_mutex);
688 tracepoint_update_probe_range(mod->tracepoints_ptrs,
689 mod->tracepoints_ptrs + mod->num_tracepoints);
690 list_for_each_entry(pos, &tracepoint_module_list, list) {
691 if (pos->tracepoints_ptrs == mod->tracepoints_ptrs) {
692 list_del(&pos->list);
698 * In the case of modules that were tainted at "coming", we'll simply
699 * walk through the list without finding it. We cannot use the "tainted"
700 * flag on "going", in case a module taints the kernel only after being
703 mutex_unlock(&tracepoints_mutex);
707 int tracepoint_module_notify(struct notifier_block *self,
708 unsigned long val, void *data)
710 struct module *mod = data;
714 case MODULE_STATE_COMING:
715 ret = tracepoint_module_coming(mod);
717 case MODULE_STATE_LIVE:
719 case MODULE_STATE_GOING:
720 ret = tracepoint_module_going(mod);
726 struct notifier_block tracepoint_module_nb = {
727 .notifier_call = tracepoint_module_notify,
731 static int init_tracepoints(void)
733 return register_module_notifier(&tracepoint_module_nb);
735 __initcall(init_tracepoints);
736 #endif /* CONFIG_MODULES */
738 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
740 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
741 static int sys_tracepoint_refcount;
743 void syscall_regfunc(void)
746 struct task_struct *g, *t;
748 if (!sys_tracepoint_refcount) {
749 read_lock_irqsave(&tasklist_lock, flags);
750 do_each_thread(g, t) {
751 /* Skip kernel threads. */
753 set_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
754 } while_each_thread(g, t);
755 read_unlock_irqrestore(&tasklist_lock, flags);
757 sys_tracepoint_refcount++;
760 void syscall_unregfunc(void)
763 struct task_struct *g, *t;
765 sys_tracepoint_refcount--;
766 if (!sys_tracepoint_refcount) {
767 read_lock_irqsave(&tasklist_lock, flags);
768 do_each_thread(g, t) {
769 clear_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
770 } while_each_thread(g, t);
771 read_unlock_irqrestore(&tasklist_lock, flags);