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. */
65 int enabled; /* Tracepoint enabled */
71 struct tracepoint_func probes[0];
74 static inline void *allocate_probes(int count)
76 struct tp_probes *p = kmalloc(count * sizeof(struct tracepoint_func)
77 + sizeof(struct tp_probes), GFP_KERNEL);
78 return p == NULL ? NULL : p->probes;
81 static void rcu_free_old_probes(struct rcu_head *head)
83 kfree(container_of(head, struct tp_probes, rcu));
86 static inline void release_probes(struct tracepoint_func *old)
89 struct tp_probes *tp_probes = container_of(old,
90 struct tp_probes, probes[0]);
91 call_rcu_sched(&tp_probes->rcu, rcu_free_old_probes);
95 static void debug_print_probes(struct tracepoint_entry *entry)
99 if (!tracepoint_debug || !entry->funcs)
102 for (i = 0; entry->funcs[i].func; i++)
103 printk(KERN_DEBUG "Probe %d : %p\n", i, entry->funcs[i].func);
106 static struct tracepoint_func *
107 tracepoint_entry_add_probe(struct tracepoint_entry *entry,
108 void *probe, void *data)
111 struct tracepoint_func *old, *new;
114 return ERR_PTR(-EINVAL);
116 debug_print_probes(entry);
119 /* (N -> N+1), (N != 0, 1) probes */
120 for (nr_probes = 0; old[nr_probes].func; nr_probes++)
121 if (old[nr_probes].func == probe &&
122 old[nr_probes].data == data)
123 return ERR_PTR(-EEXIST);
125 /* + 2 : one for new probe, one for NULL func */
126 new = allocate_probes(nr_probes + 2);
128 return ERR_PTR(-ENOMEM);
130 memcpy(new, old, nr_probes * sizeof(struct tracepoint_func));
131 new[nr_probes].func = probe;
132 new[nr_probes].data = data;
133 new[nr_probes + 1].func = NULL;
134 entry->refcount = nr_probes + 1;
136 debug_print_probes(entry);
141 tracepoint_entry_remove_probe(struct tracepoint_entry *entry,
142 void *probe, void *data)
144 int nr_probes = 0, nr_del = 0, i;
145 struct tracepoint_func *old, *new;
150 return ERR_PTR(-ENOENT);
152 debug_print_probes(entry);
153 /* (N -> M), (N > 1, M >= 0) probes */
155 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
156 if (old[nr_probes].func == probe &&
157 old[nr_probes].data == data)
163 * If probe is NULL, then nr_probes = nr_del = 0, and then the
164 * entire entry will be removed.
166 if (nr_probes - nr_del == 0) {
167 /* N -> 0, (N > 1) */
170 debug_print_probes(entry);
174 /* N -> M, (N > 1, M > 0) */
176 new = allocate_probes(nr_probes - nr_del + 1);
178 return ERR_PTR(-ENOMEM);
179 for (i = 0; old[i].func; i++)
180 if (old[i].func != probe || old[i].data != data)
182 new[nr_probes - nr_del].func = NULL;
183 entry->refcount = nr_probes - nr_del;
186 debug_print_probes(entry);
191 * Get tracepoint if the tracepoint is present in the tracepoint hash table.
192 * Must be called with tracepoints_mutex held.
193 * Returns NULL if not present.
195 static struct tracepoint_entry *get_tracepoint(const char *name)
197 struct hlist_head *head;
198 struct tracepoint_entry *e;
199 u32 hash = jhash(name, strlen(name), 0);
201 head = &tracepoint_table[hash & (TRACEPOINT_TABLE_SIZE - 1)];
202 hlist_for_each_entry(e, head, hlist) {
203 if (!strcmp(name, e->name))
210 * Add the tracepoint to the tracepoint hash table. Must be called with
211 * tracepoints_mutex held.
213 static struct tracepoint_entry *add_tracepoint(const char *name)
215 struct hlist_head *head;
216 struct tracepoint_entry *e;
217 size_t name_len = strlen(name) + 1;
218 u32 hash = jhash(name, name_len-1, 0);
220 head = &tracepoint_table[hash & (TRACEPOINT_TABLE_SIZE - 1)];
221 hlist_for_each_entry(e, head, hlist) {
222 if (!strcmp(name, e->name)) {
224 "tracepoint %s busy\n", name);
225 return ERR_PTR(-EEXIST); /* Already there */
229 * Using kmalloc here to allocate a variable length element. Could
230 * cause some memory fragmentation if overused.
232 e = kmalloc(sizeof(struct tracepoint_entry) + name_len, GFP_KERNEL);
234 return ERR_PTR(-ENOMEM);
235 memcpy(&e->name[0], name, name_len);
239 hlist_add_head(&e->hlist, head);
244 * Remove the tracepoint from the tracepoint hash table. Must be called with
247 static inline void remove_tracepoint(struct tracepoint_entry *e)
249 hlist_del(&e->hlist);
254 * Sets the probe callback corresponding to one tracepoint.
256 static void set_tracepoint(struct tracepoint_entry **entry,
257 struct tracepoint *elem, int active)
259 WARN_ON(strcmp((*entry)->name, elem->name) != 0);
261 if (elem->regfunc && !static_key_enabled(&elem->key) && active)
263 else if (elem->unregfunc && static_key_enabled(&elem->key) && !active)
267 * rcu_assign_pointer has a smp_wmb() which makes sure that the new
268 * probe callbacks array is consistent before setting a pointer to it.
269 * This array is referenced by __DO_TRACE from
270 * include/linux/tracepoints.h. A matching smp_read_barrier_depends()
273 rcu_assign_pointer(elem->funcs, (*entry)->funcs);
274 if (active && !static_key_enabled(&elem->key))
275 static_key_slow_inc(&elem->key);
276 else if (!active && static_key_enabled(&elem->key))
277 static_key_slow_dec(&elem->key);
281 * Disable a tracepoint and its probe callback.
282 * Note: only waiting an RCU period after setting elem->call to the empty
283 * function insures that the original callback is not used anymore. This insured
284 * by preempt_disable around the call site.
286 static void disable_tracepoint(struct tracepoint *elem)
288 if (elem->unregfunc && static_key_enabled(&elem->key))
291 if (static_key_enabled(&elem->key))
292 static_key_slow_dec(&elem->key);
293 rcu_assign_pointer(elem->funcs, NULL);
297 * tracepoint_update_probe_range - Update a probe range
298 * @begin: beginning of the range
299 * @end: end of the range
301 * Updates the probe callback corresponding to a range of tracepoints.
302 * Called with tracepoints_mutex held.
304 static void tracepoint_update_probe_range(struct tracepoint * const *begin,
305 struct tracepoint * const *end)
307 struct tracepoint * const *iter;
308 struct tracepoint_entry *mark_entry;
313 for (iter = begin; iter < end; iter++) {
314 mark_entry = get_tracepoint((*iter)->name);
316 set_tracepoint(&mark_entry, *iter,
317 !!mark_entry->refcount);
318 mark_entry->enabled = !!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
376 * @data: probe private data
379 * - 0 if the probe was successfully registered, and tracepoint
380 * callsites are currently loaded for that probe,
381 * - -ENODEV if the probe was successfully registered, but no tracepoint
382 * callsite is currently loaded for that probe,
383 * - other negative error value on error.
385 * When tracepoint_probe_register() returns either 0 or -ENODEV,
386 * parameters @name, @probe, and @data may be used by the tracepoint
387 * infrastructure until the probe is unregistered.
389 * The probe address must at least be aligned on the architecture pointer size.
391 int tracepoint_probe_register(const char *name, void *probe, void *data)
393 struct tracepoint_func *old;
394 struct tracepoint_entry *entry;
397 mutex_lock(&tracepoints_mutex);
398 old = tracepoint_add_probe(name, probe, data);
400 mutex_unlock(&tracepoints_mutex);
403 tracepoint_update_probes(); /* may update entry */
404 entry = get_tracepoint(name);
405 /* Make sure the entry was enabled */
406 if (!entry || !entry->enabled)
408 mutex_unlock(&tracepoints_mutex);
412 EXPORT_SYMBOL_GPL(tracepoint_probe_register);
414 static struct tracepoint_func *
415 tracepoint_remove_probe(const char *name, void *probe, void *data)
417 struct tracepoint_entry *entry;
418 struct tracepoint_func *old;
420 entry = get_tracepoint(name);
422 return ERR_PTR(-ENOENT);
423 old = tracepoint_entry_remove_probe(entry, probe, data);
426 if (!entry->refcount)
427 remove_tracepoint(entry);
432 * tracepoint_probe_unregister - Disconnect a probe from a tracepoint
433 * @name: tracepoint name
434 * @probe: probe function pointer
435 * @data: probe private data
437 * We do not need to call a synchronize_sched to make sure the probes have
438 * finished running before doing a module unload, because the module unload
439 * itself uses stop_machine(), which insures that every preempt disabled section
442 int tracepoint_probe_unregister(const char *name, void *probe, void *data)
444 struct tracepoint_func *old;
446 mutex_lock(&tracepoints_mutex);
447 old = tracepoint_remove_probe(name, probe, data);
449 mutex_unlock(&tracepoints_mutex);
452 tracepoint_update_probes(); /* may update entry */
453 mutex_unlock(&tracepoints_mutex);
457 EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
460 #ifdef CONFIG_MODULES
461 bool trace_module_has_bad_taint(struct module *mod)
463 return mod->taints & ~((1 << TAINT_OOT_MODULE) | (1 << TAINT_CRAP));
466 static int tracepoint_module_coming(struct module *mod)
468 struct tp_module *tp_mod;
471 if (!mod->num_tracepoints)
475 * We skip modules that taint the kernel, especially those with different
476 * module headers (for forced load), to make sure we don't cause a crash.
477 * Staging and out-of-tree GPL modules are fine.
479 if (trace_module_has_bad_taint(mod))
481 mutex_lock(&tracepoints_mutex);
482 tp_mod = kmalloc(sizeof(struct tp_module), GFP_KERNEL);
487 tp_mod->num_tracepoints = mod->num_tracepoints;
488 tp_mod->tracepoints_ptrs = mod->tracepoints_ptrs;
489 list_add_tail(&tp_mod->list, &tracepoint_module_list);
490 tracepoint_update_probe_range(mod->tracepoints_ptrs,
491 mod->tracepoints_ptrs + mod->num_tracepoints);
493 mutex_unlock(&tracepoints_mutex);
497 static int tracepoint_module_going(struct module *mod)
499 struct tp_module *pos;
501 if (!mod->num_tracepoints)
504 mutex_lock(&tracepoints_mutex);
505 tracepoint_update_probe_range(mod->tracepoints_ptrs,
506 mod->tracepoints_ptrs + mod->num_tracepoints);
507 list_for_each_entry(pos, &tracepoint_module_list, list) {
508 if (pos->tracepoints_ptrs == mod->tracepoints_ptrs) {
509 list_del(&pos->list);
515 * In the case of modules that were tainted at "coming", we'll simply
516 * walk through the list without finding it. We cannot use the "tainted"
517 * flag on "going", in case a module taints the kernel only after being
520 mutex_unlock(&tracepoints_mutex);
524 int tracepoint_module_notify(struct notifier_block *self,
525 unsigned long val, void *data)
527 struct module *mod = data;
531 case MODULE_STATE_COMING:
532 ret = tracepoint_module_coming(mod);
534 case MODULE_STATE_LIVE:
536 case MODULE_STATE_GOING:
537 ret = tracepoint_module_going(mod);
543 struct notifier_block tracepoint_module_nb = {
544 .notifier_call = tracepoint_module_notify,
548 static int init_tracepoints(void)
550 return register_module_notifier(&tracepoint_module_nb);
552 __initcall(init_tracepoints);
553 #endif /* CONFIG_MODULES */
555 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
557 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
558 static int sys_tracepoint_refcount;
560 void syscall_regfunc(void)
563 struct task_struct *g, *t;
565 if (!sys_tracepoint_refcount) {
566 read_lock_irqsave(&tasklist_lock, flags);
567 do_each_thread(g, t) {
568 /* Skip kernel threads. */
570 set_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
571 } while_each_thread(g, t);
572 read_unlock_irqrestore(&tasklist_lock, flags);
574 sys_tracepoint_refcount++;
577 void syscall_unregfunc(void)
580 struct task_struct *g, *t;
582 sys_tracepoint_refcount--;
583 if (!sys_tracepoint_refcount) {
584 read_lock_irqsave(&tasklist_lock, flags);
585 do_each_thread(g, t) {
586 clear_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
587 } while_each_thread(g, t);
588 read_unlock_irqrestore(&tasklist_lock, flags);