xref: /linux/kernel/tracepoint.c (revision 323bbfcf1ef8836d0d2ad9e2c1f1c684f0e3b5b3)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (C) 2008-2014 Mathieu Desnoyers
4  */
5 #include <linux/module.h>
6 #include <linux/mutex.h>
7 #include <linux/types.h>
8 #include <linux/jhash.h>
9 #include <linux/list.h>
10 #include <linux/rcupdate.h>
11 #include <linux/tracepoint.h>
12 #include <linux/err.h>
13 #include <linux/slab.h>
14 #include <linux/sched/signal.h>
15 #include <linux/sched/task.h>
16 #include <linux/static_key.h>
17 
18 enum tp_func_state {
19 	TP_FUNC_0,
20 	TP_FUNC_1,
21 	TP_FUNC_2,
22 	TP_FUNC_N,
23 };
24 
25 extern tracepoint_ptr_t __start___tracepoints_ptrs[];
26 extern tracepoint_ptr_t __stop___tracepoints_ptrs[];
27 
28 enum tp_transition_sync {
29 	TP_TRANSITION_SYNC_1_0_1,
30 	TP_TRANSITION_SYNC_N_2_1,
31 
32 	_NR_TP_TRANSITION_SYNC,
33 };
34 
35 struct tp_transition_snapshot {
36 	unsigned long rcu;
37 	unsigned long srcu_gp;
38 	bool ongoing;
39 };
40 
41 DEFINE_SRCU_FAST(tracepoint_srcu);
42 EXPORT_SYMBOL_GPL(tracepoint_srcu);
43 
44 /* Protected by tracepoints_mutex */
45 static struct tp_transition_snapshot tp_transition_snapshot[_NR_TP_TRANSITION_SYNC];
46 
47 static void tp_rcu_get_state(enum tp_transition_sync sync)
48 {
49 	struct tp_transition_snapshot *snapshot = &tp_transition_snapshot[sync];
50 
51 	/* Keep the latest get_state snapshot. */
52 	snapshot->rcu = get_state_synchronize_rcu();
53 	snapshot->srcu_gp = start_poll_synchronize_srcu(&tracepoint_srcu);
54 	snapshot->ongoing = true;
55 }
56 
57 static void tp_rcu_cond_sync(enum tp_transition_sync sync)
58 {
59 	struct tp_transition_snapshot *snapshot = &tp_transition_snapshot[sync];
60 
61 	if (!snapshot->ongoing)
62 		return;
63 	cond_synchronize_rcu(snapshot->rcu);
64 	if (!poll_state_synchronize_srcu(&tracepoint_srcu, snapshot->srcu_gp))
65 		synchronize_srcu(&tracepoint_srcu);
66 	snapshot->ongoing = false;
67 }
68 
69 /* Set to 1 to enable tracepoint debug output */
70 static const int tracepoint_debug;
71 
72 #ifdef CONFIG_MODULES
73 /*
74  * Tracepoint module list mutex protects the local module list.
75  */
76 static DEFINE_MUTEX(tracepoint_module_list_mutex);
77 
78 /* Local list of struct tp_module */
79 static LIST_HEAD(tracepoint_module_list);
80 #endif /* CONFIG_MODULES */
81 
82 /*
83  * tracepoints_mutex protects the builtin and module tracepoints.
84  * tracepoints_mutex nests inside tracepoint_module_list_mutex.
85  */
86 static DEFINE_MUTEX(tracepoints_mutex);
87 
88 /*
89  * Note about RCU :
90  * It is used to delay the free of multiple probes array until a quiescent
91  * state is reached.
92  */
93 struct tp_probes {
94 	struct rcu_head rcu;
95 	struct tracepoint_func probes[];
96 };
97 
98 /* Called in removal of a func but failed to allocate a new tp_funcs */
99 static void tp_stub_func(void)
100 {
101 	return;
102 }
103 
104 static inline void *allocate_probes(int count)
105 {
106 	struct tp_probes *p  = kmalloc_flex(*p, probes, count);
107 	return p == NULL ? NULL : p->probes;
108 }
109 
110 static void rcu_free_old_probes(struct rcu_head *head)
111 {
112 	kfree(container_of(head, struct tp_probes, rcu));
113 }
114 
115 static inline void release_probes(struct tracepoint *tp, struct tracepoint_func *old)
116 {
117 	if (old) {
118 		struct tp_probes *tp_probes = container_of(old,
119 			struct tp_probes, probes[0]);
120 
121 		if (tracepoint_is_faultable(tp)) {
122 			call_rcu_tasks_trace(&tp_probes->rcu,
123 					     rcu_free_old_probes);
124 		} else {
125 			call_srcu(&tracepoint_srcu, &tp_probes->rcu,
126 				  rcu_free_old_probes);
127 		}
128 	}
129 }
130 
131 static void debug_print_probes(struct tracepoint_func *funcs)
132 {
133 	int i;
134 
135 	if (!tracepoint_debug || !funcs)
136 		return;
137 
138 	for (i = 0; funcs[i].func; i++)
139 		printk(KERN_DEBUG "Probe %d : %pSb\n", i, funcs[i].func);
140 }
141 
142 static struct tracepoint_func *
143 func_add(struct tracepoint_func **funcs, struct tracepoint_func *tp_func,
144 	 int prio)
145 {
146 	struct tracepoint_func *old, *new;
147 	int iter_probes;	/* Iterate over old probe array. */
148 	int nr_probes = 0;	/* Counter for probes */
149 	int pos = -1;		/* Insertion position into new array */
150 
151 	if (WARN_ON(!tp_func->func))
152 		return ERR_PTR(-EINVAL);
153 
154 	debug_print_probes(*funcs);
155 	old = *funcs;
156 	if (old) {
157 		/* (N -> N+1), (N != 0, 1) probes */
158 		for (iter_probes = 0; old[iter_probes].func; iter_probes++) {
159 			if (old[iter_probes].func == tp_stub_func)
160 				continue;	/* Skip stub functions. */
161 			if (old[iter_probes].func == tp_func->func &&
162 			    old[iter_probes].data == tp_func->data)
163 				return ERR_PTR(-EEXIST);
164 			nr_probes++;
165 		}
166 	}
167 	/* + 2 : one for new probe, one for NULL func */
168 	new = allocate_probes(nr_probes + 2);
169 	if (new == NULL)
170 		return ERR_PTR(-ENOMEM);
171 	if (old) {
172 		nr_probes = 0;
173 		for (iter_probes = 0; old[iter_probes].func; iter_probes++) {
174 			if (old[iter_probes].func == tp_stub_func)
175 				continue;
176 			/* Insert before probes of lower priority */
177 			if (pos < 0 && old[iter_probes].prio < prio)
178 				pos = nr_probes++;
179 			new[nr_probes++] = old[iter_probes];
180 		}
181 		if (pos < 0)
182 			pos = nr_probes++;
183 		/* nr_probes now points to the end of the new array */
184 	} else {
185 		pos = 0;
186 		nr_probes = 1; /* must point at end of array */
187 	}
188 	new[pos] = *tp_func;
189 	new[nr_probes].func = NULL;
190 	*funcs = new;
191 	debug_print_probes(*funcs);
192 	return old;
193 }
194 
195 static void *func_remove(struct tracepoint_func **funcs,
196 		struct tracepoint_func *tp_func)
197 {
198 	int nr_probes = 0, nr_del = 0, i;
199 	struct tracepoint_func *old, *new;
200 
201 	old = *funcs;
202 
203 	if (!old)
204 		return ERR_PTR(-ENOENT);
205 
206 	debug_print_probes(*funcs);
207 	/* (N -> M), (N > 1, M >= 0) probes */
208 	if (tp_func->func) {
209 		for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
210 			if ((old[nr_probes].func == tp_func->func &&
211 			     old[nr_probes].data == tp_func->data) ||
212 			    old[nr_probes].func == tp_stub_func)
213 				nr_del++;
214 		}
215 	}
216 
217 	/*
218 	 * If probe is NULL, then nr_probes = nr_del = 0, and then the
219 	 * entire entry will be removed.
220 	 */
221 	if (nr_probes - nr_del == 0) {
222 		/* N -> 0, (N > 1) */
223 		*funcs = NULL;
224 		debug_print_probes(*funcs);
225 		return old;
226 	} else {
227 		int j = 0;
228 		/* N -> M, (N > 1, M > 0) */
229 		/* + 1 for NULL */
230 		new = allocate_probes(nr_probes - nr_del + 1);
231 		if (new) {
232 			for (i = 0; old[i].func; i++) {
233 				if ((old[i].func != tp_func->func ||
234 				     old[i].data != tp_func->data) &&
235 				    old[i].func != tp_stub_func)
236 					new[j++] = old[i];
237 			}
238 			new[nr_probes - nr_del].func = NULL;
239 			*funcs = new;
240 		} else {
241 			/*
242 			 * Failed to allocate, replace the old function
243 			 * with calls to tp_stub_func.
244 			 */
245 			for (i = 0; old[i].func; i++) {
246 				if (old[i].func == tp_func->func &&
247 				    old[i].data == tp_func->data)
248 					WRITE_ONCE(old[i].func, tp_stub_func);
249 			}
250 			*funcs = old;
251 		}
252 	}
253 	debug_print_probes(*funcs);
254 	return old;
255 }
256 
257 /*
258  * Count the number of functions (enum tp_func_state) in a tp_funcs array.
259  */
260 static enum tp_func_state nr_func_state(const struct tracepoint_func *tp_funcs)
261 {
262 	if (!tp_funcs)
263 		return TP_FUNC_0;
264 	if (!tp_funcs[1].func)
265 		return TP_FUNC_1;
266 	if (!tp_funcs[2].func)
267 		return TP_FUNC_2;
268 	return TP_FUNC_N;	/* 3 or more */
269 }
270 
271 static void tracepoint_update_call(struct tracepoint *tp, struct tracepoint_func *tp_funcs)
272 {
273 	void *func = tp->iterator;
274 
275 	/* Synthetic events do not have static call sites */
276 	if (!tp->static_call_key)
277 		return;
278 	if (nr_func_state(tp_funcs) == TP_FUNC_1)
279 		func = tp_funcs[0].func;
280 	__static_call_update(tp->static_call_key, tp->static_call_tramp, func);
281 }
282 
283 /*
284  * Add the probe function to a tracepoint.
285  */
286 static int tracepoint_add_func(struct tracepoint *tp,
287 			       struct tracepoint_func *func, int prio,
288 			       bool warn)
289 {
290 	struct tracepoint_func *old, *tp_funcs;
291 	int ret;
292 
293 	if (tp->ext && tp->ext->regfunc && !static_key_enabled(&tp->key)) {
294 		ret = tp->ext->regfunc();
295 		if (ret < 0)
296 			return ret;
297 	}
298 
299 	tp_funcs = rcu_dereference_protected(tp->funcs,
300 			lockdep_is_held(&tracepoints_mutex));
301 	old = func_add(&tp_funcs, func, prio);
302 	if (IS_ERR(old)) {
303 		WARN_ON_ONCE(warn && PTR_ERR(old) != -ENOMEM);
304 		return PTR_ERR(old);
305 	}
306 
307 	/*
308 	 * rcu_assign_pointer has as smp_store_release() which makes sure
309 	 * that the new probe callbacks array is consistent before setting
310 	 * a pointer to it.  This array is referenced by __DO_TRACE from
311 	 * include/linux/tracepoint.h using rcu_dereference_sched().
312 	 */
313 	switch (nr_func_state(tp_funcs)) {
314 	case TP_FUNC_1:		/* 0->1 */
315 		/*
316 		 * Make sure new static func never uses old data after a
317 		 * 1->0->1 transition sequence.
318 		 */
319 		tp_rcu_cond_sync(TP_TRANSITION_SYNC_1_0_1);
320 		/* Set static call to first function */
321 		tracepoint_update_call(tp, tp_funcs);
322 		/* Both iterator and static call handle NULL tp->funcs */
323 		rcu_assign_pointer(tp->funcs, tp_funcs);
324 		static_branch_enable(&tp->key);
325 		break;
326 	case TP_FUNC_2:		/* 1->2 */
327 		/* Set iterator static call */
328 		tracepoint_update_call(tp, tp_funcs);
329 		/*
330 		 * Iterator callback installed before updating tp->funcs.
331 		 * Requires ordering between RCU assign/dereference and
332 		 * static call update/call.
333 		 */
334 		fallthrough;
335 	case TP_FUNC_N:		/* N->N+1 (N>1) */
336 		rcu_assign_pointer(tp->funcs, tp_funcs);
337 		/*
338 		 * Make sure static func never uses incorrect data after a
339 		 * N->...->2->1 (N>1) transition sequence.
340 		 */
341 		if (tp_funcs[0].data != old[0].data)
342 			tp_rcu_get_state(TP_TRANSITION_SYNC_N_2_1);
343 		break;
344 	default:
345 		WARN_ON_ONCE(1);
346 		break;
347 	}
348 
349 	release_probes(tp, old);
350 	return 0;
351 }
352 
353 /*
354  * Remove a probe function from a tracepoint.
355  * Note: only waiting an RCU period after setting elem->call to the empty
356  * function insures that the original callback is not used anymore. This insured
357  * by preempt_disable around the call site.
358  */
359 static int tracepoint_remove_func(struct tracepoint *tp,
360 		struct tracepoint_func *func)
361 {
362 	struct tracepoint_func *old, *tp_funcs;
363 
364 	tp_funcs = rcu_dereference_protected(tp->funcs,
365 			lockdep_is_held(&tracepoints_mutex));
366 	old = func_remove(&tp_funcs, func);
367 	if (WARN_ON_ONCE(IS_ERR(old)))
368 		return PTR_ERR(old);
369 
370 	if (tp_funcs == old)
371 		/* Failed allocating new tp_funcs, replaced func with stub */
372 		return 0;
373 
374 	switch (nr_func_state(tp_funcs)) {
375 	case TP_FUNC_0:		/* 1->0 */
376 		/* Removed last function */
377 		if (tp->ext && tp->ext->unregfunc && static_key_enabled(&tp->key))
378 			tp->ext->unregfunc();
379 		static_branch_disable(&tp->key);
380 		/* Set iterator static call */
381 		tracepoint_update_call(tp, tp_funcs);
382 		/* Both iterator and static call handle NULL tp->funcs */
383 		rcu_assign_pointer(tp->funcs, NULL);
384 		/*
385 		 * Make sure new static func never uses old data after a
386 		 * 1->0->1 transition sequence.
387 		 */
388 		tp_rcu_get_state(TP_TRANSITION_SYNC_1_0_1);
389 		break;
390 	case TP_FUNC_1:		/* 2->1 */
391 		rcu_assign_pointer(tp->funcs, tp_funcs);
392 		/*
393 		 * Make sure static func never uses incorrect data after a
394 		 * N->...->2->1 (N>2) transition sequence. If the first
395 		 * element's data has changed, then force the synchronization
396 		 * to prevent current readers that have loaded the old data
397 		 * from calling the new function.
398 		 */
399 		if (tp_funcs[0].data != old[0].data)
400 			tp_rcu_get_state(TP_TRANSITION_SYNC_N_2_1);
401 		tp_rcu_cond_sync(TP_TRANSITION_SYNC_N_2_1);
402 		/* Set static call to first function */
403 		tracepoint_update_call(tp, tp_funcs);
404 		break;
405 	case TP_FUNC_2:		/* N->N-1 (N>2) */
406 		fallthrough;
407 	case TP_FUNC_N:
408 		rcu_assign_pointer(tp->funcs, tp_funcs);
409 		/*
410 		 * Make sure static func never uses incorrect data after a
411 		 * N->...->2->1 (N>2) transition sequence.
412 		 */
413 		if (tp_funcs[0].data != old[0].data)
414 			tp_rcu_get_state(TP_TRANSITION_SYNC_N_2_1);
415 		break;
416 	default:
417 		WARN_ON_ONCE(1);
418 		break;
419 	}
420 	release_probes(tp, old);
421 	return 0;
422 }
423 
424 /**
425  * tracepoint_probe_register_prio_may_exist -  Connect a probe to a tracepoint with priority
426  * @tp: tracepoint
427  * @probe: probe handler
428  * @data: tracepoint data
429  * @prio: priority of this function over other registered functions
430  *
431  * Same as tracepoint_probe_register_prio() except that it will not warn
432  * if the tracepoint is already registered.
433  */
434 int tracepoint_probe_register_prio_may_exist(struct tracepoint *tp, void *probe,
435 					     void *data, int prio)
436 {
437 	struct tracepoint_func tp_func;
438 	int ret;
439 
440 	mutex_lock(&tracepoints_mutex);
441 	tp_func.func = probe;
442 	tp_func.data = data;
443 	tp_func.prio = prio;
444 	ret = tracepoint_add_func(tp, &tp_func, prio, false);
445 	mutex_unlock(&tracepoints_mutex);
446 	return ret;
447 }
448 EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio_may_exist);
449 
450 /**
451  * tracepoint_probe_register_prio -  Connect a probe to a tracepoint with priority
452  * @tp: tracepoint
453  * @probe: probe handler
454  * @data: tracepoint data
455  * @prio: priority of this function over other registered functions
456  *
457  * Returns 0 if ok, error value on error.
458  * Note: if @tp is within a module, the caller is responsible for
459  * unregistering the probe before the module is gone. This can be
460  * performed either with a tracepoint module going notifier, or from
461  * within module exit functions.
462  */
463 int tracepoint_probe_register_prio(struct tracepoint *tp, void *probe,
464 				   void *data, int prio)
465 {
466 	struct tracepoint_func tp_func;
467 	int ret;
468 
469 	mutex_lock(&tracepoints_mutex);
470 	tp_func.func = probe;
471 	tp_func.data = data;
472 	tp_func.prio = prio;
473 	ret = tracepoint_add_func(tp, &tp_func, prio, true);
474 	mutex_unlock(&tracepoints_mutex);
475 	return ret;
476 }
477 EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio);
478 
479 /**
480  * tracepoint_probe_register -  Connect a probe to a tracepoint
481  * @tp: tracepoint
482  * @probe: probe handler
483  * @data: tracepoint data
484  *
485  * Returns 0 if ok, error value on error.
486  * Note: if @tp is within a module, the caller is responsible for
487  * unregistering the probe before the module is gone. This can be
488  * performed either with a tracepoint module going notifier, or from
489  * within module exit functions.
490  */
491 int tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data)
492 {
493 	return tracepoint_probe_register_prio(tp, probe, data, TRACEPOINT_DEFAULT_PRIO);
494 }
495 EXPORT_SYMBOL_GPL(tracepoint_probe_register);
496 
497 /**
498  * tracepoint_probe_unregister -  Disconnect a probe from a tracepoint
499  * @tp: tracepoint
500  * @probe: probe function pointer
501  * @data: tracepoint data
502  *
503  * Returns 0 if ok, error value on error.
504  */
505 int tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data)
506 {
507 	struct tracepoint_func tp_func;
508 	int ret;
509 
510 	mutex_lock(&tracepoints_mutex);
511 	tp_func.func = probe;
512 	tp_func.data = data;
513 	ret = tracepoint_remove_func(tp, &tp_func);
514 	mutex_unlock(&tracepoints_mutex);
515 	return ret;
516 }
517 EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
518 
519 static void for_each_tracepoint_range(
520 		tracepoint_ptr_t *begin, tracepoint_ptr_t *end,
521 		void (*fct)(struct tracepoint *tp, void *priv),
522 		void *priv)
523 {
524 	tracepoint_ptr_t *iter;
525 
526 	if (!begin)
527 		return;
528 	for (iter = begin; iter < end; iter++)
529 		fct(tracepoint_ptr_deref(iter), priv);
530 }
531 
532 #ifdef CONFIG_MODULES
533 bool trace_module_has_bad_taint(struct module *mod)
534 {
535 	return mod->taints & ~((1 << TAINT_OOT_MODULE) | (1 << TAINT_CRAP) |
536 				(1 << TAINT_UNSIGNED_MODULE) | (1 << TAINT_TEST) |
537 				(1 << TAINT_LIVEPATCH));
538 }
539 
540 static BLOCKING_NOTIFIER_HEAD(tracepoint_notify_list);
541 
542 /**
543  * register_tracepoint_module_notifier - register tracepoint coming/going notifier
544  * @nb: notifier block
545  *
546  * Notifiers registered with this function are called on module
547  * coming/going with the tracepoint_module_list_mutex held.
548  * The notifier block callback should expect a "struct tp_module" data
549  * pointer.
550  */
551 int register_tracepoint_module_notifier(struct notifier_block *nb)
552 {
553 	struct tp_module *tp_mod;
554 	int ret;
555 
556 	mutex_lock(&tracepoint_module_list_mutex);
557 	ret = blocking_notifier_chain_register(&tracepoint_notify_list, nb);
558 	if (ret)
559 		goto end;
560 	list_for_each_entry(tp_mod, &tracepoint_module_list, list)
561 		(void) nb->notifier_call(nb, MODULE_STATE_COMING, tp_mod);
562 end:
563 	mutex_unlock(&tracepoint_module_list_mutex);
564 	return ret;
565 }
566 EXPORT_SYMBOL_GPL(register_tracepoint_module_notifier);
567 
568 /**
569  * unregister_tracepoint_module_notifier - unregister tracepoint coming/going notifier
570  * @nb: notifier block
571  *
572  * The notifier block callback should expect a "struct tp_module" data
573  * pointer.
574  */
575 int unregister_tracepoint_module_notifier(struct notifier_block *nb)
576 {
577 	struct tp_module *tp_mod;
578 	int ret;
579 
580 	mutex_lock(&tracepoint_module_list_mutex);
581 	ret = blocking_notifier_chain_unregister(&tracepoint_notify_list, nb);
582 	if (ret)
583 		goto end;
584 	list_for_each_entry(tp_mod, &tracepoint_module_list, list)
585 		(void) nb->notifier_call(nb, MODULE_STATE_GOING, tp_mod);
586 end:
587 	mutex_unlock(&tracepoint_module_list_mutex);
588 	return ret;
589 
590 }
591 EXPORT_SYMBOL_GPL(unregister_tracepoint_module_notifier);
592 
593 /*
594  * Ensure the tracer unregistered the module's probes before the module
595  * teardown is performed. Prevents leaks of probe and data pointers.
596  */
597 static void tp_module_going_check_quiescent(struct tracepoint *tp, void *priv)
598 {
599 	WARN_ON_ONCE(tp->funcs);
600 }
601 
602 static int tracepoint_module_coming(struct module *mod)
603 {
604 	struct tp_module *tp_mod;
605 
606 	if (!mod->num_tracepoints)
607 		return 0;
608 
609 	/*
610 	 * We skip modules that taint the kernel, especially those with different
611 	 * module headers (for forced load), to make sure we don't cause a crash.
612 	 * Staging, out-of-tree, unsigned GPL, and test modules are fine.
613 	 */
614 	if (trace_module_has_bad_taint(mod))
615 		return 0;
616 
617 	tp_mod = kmalloc_obj(struct tp_module);
618 	if (!tp_mod)
619 		return -ENOMEM;
620 	tp_mod->mod = mod;
621 
622 	mutex_lock(&tracepoint_module_list_mutex);
623 	list_add_tail(&tp_mod->list, &tracepoint_module_list);
624 	blocking_notifier_call_chain(&tracepoint_notify_list,
625 			MODULE_STATE_COMING, tp_mod);
626 	mutex_unlock(&tracepoint_module_list_mutex);
627 	return 0;
628 }
629 
630 static void tracepoint_module_going(struct module *mod)
631 {
632 	struct tp_module *tp_mod;
633 
634 	if (!mod->num_tracepoints)
635 		return;
636 
637 	mutex_lock(&tracepoint_module_list_mutex);
638 	list_for_each_entry(tp_mod, &tracepoint_module_list, list) {
639 		if (tp_mod->mod == mod) {
640 			blocking_notifier_call_chain(&tracepoint_notify_list,
641 					MODULE_STATE_GOING, tp_mod);
642 			list_del(&tp_mod->list);
643 			kfree(tp_mod);
644 			/*
645 			 * Called the going notifier before checking for
646 			 * quiescence.
647 			 */
648 			for_each_tracepoint_range(mod->tracepoints_ptrs,
649 				mod->tracepoints_ptrs + mod->num_tracepoints,
650 				tp_module_going_check_quiescent, NULL);
651 			break;
652 		}
653 	}
654 	/*
655 	 * In the case of modules that were tainted at "coming", we'll simply
656 	 * walk through the list without finding it. We cannot use the "tainted"
657 	 * flag on "going", in case a module taints the kernel only after being
658 	 * loaded.
659 	 */
660 	mutex_unlock(&tracepoint_module_list_mutex);
661 }
662 
663 static int tracepoint_module_notify(struct notifier_block *self,
664 		unsigned long val, void *data)
665 {
666 	struct module *mod = data;
667 	int ret = 0;
668 
669 	switch (val) {
670 	case MODULE_STATE_COMING:
671 		ret = tracepoint_module_coming(mod);
672 		break;
673 	case MODULE_STATE_LIVE:
674 		break;
675 	case MODULE_STATE_GOING:
676 		tracepoint_module_going(mod);
677 		break;
678 	case MODULE_STATE_UNFORMED:
679 		break;
680 	}
681 	return notifier_from_errno(ret);
682 }
683 
684 static struct notifier_block tracepoint_module_nb = {
685 	.notifier_call = tracepoint_module_notify,
686 	.priority = 0,
687 };
688 
689 static __init int init_tracepoints(void)
690 {
691 	int ret;
692 
693 	ret = register_module_notifier(&tracepoint_module_nb);
694 	if (ret)
695 		pr_warn("Failed to register tracepoint module enter notifier\n");
696 
697 	return ret;
698 }
699 __initcall(init_tracepoints);
700 
701 /**
702  * for_each_tracepoint_in_module - iteration on all tracepoints in a module
703  * @mod: module
704  * @fct: callback
705  * @priv: private data
706  */
707 void for_each_tracepoint_in_module(struct module *mod,
708 				   void (*fct)(struct tracepoint *tp,
709 				    struct module *mod, void *priv),
710 				   void *priv)
711 {
712 	tracepoint_ptr_t *begin, *end, *iter;
713 
714 	lockdep_assert_held(&tracepoint_module_list_mutex);
715 
716 	if (!mod)
717 		return;
718 
719 	begin = mod->tracepoints_ptrs;
720 	end = mod->tracepoints_ptrs + mod->num_tracepoints;
721 
722 	for (iter = begin; iter < end; iter++)
723 		fct(tracepoint_ptr_deref(iter), mod, priv);
724 }
725 
726 /**
727  * for_each_module_tracepoint - iteration on all tracepoints in all modules
728  * @fct: callback
729  * @priv: private data
730  */
731 void for_each_module_tracepoint(void (*fct)(struct tracepoint *tp,
732 				 struct module *mod, void *priv),
733 				void *priv)
734 {
735 	struct tp_module *tp_mod;
736 
737 	mutex_lock(&tracepoint_module_list_mutex);
738 	list_for_each_entry(tp_mod, &tracepoint_module_list, list)
739 		for_each_tracepoint_in_module(tp_mod->mod, fct, priv);
740 	mutex_unlock(&tracepoint_module_list_mutex);
741 }
742 #endif /* CONFIG_MODULES */
743 
744 /**
745  * for_each_kernel_tracepoint - iteration on all kernel tracepoints
746  * @fct: callback
747  * @priv: private data
748  */
749 void for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
750 		void *priv)
751 {
752 	for_each_tracepoint_range(__start___tracepoints_ptrs,
753 		__stop___tracepoints_ptrs, fct, priv);
754 }
755 EXPORT_SYMBOL_GPL(for_each_kernel_tracepoint);
756 
757 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
758 
759 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
760 static int sys_tracepoint_refcount;
761 
762 int syscall_regfunc(void)
763 {
764 	struct task_struct *p, *t;
765 
766 	if (!sys_tracepoint_refcount) {
767 		read_lock(&tasklist_lock);
768 		for_each_process_thread(p, t) {
769 			set_task_syscall_work(t, SYSCALL_TRACEPOINT);
770 		}
771 		read_unlock(&tasklist_lock);
772 	}
773 	sys_tracepoint_refcount++;
774 
775 	return 0;
776 }
777 
778 void syscall_unregfunc(void)
779 {
780 	struct task_struct *p, *t;
781 
782 	sys_tracepoint_refcount--;
783 	if (!sys_tracepoint_refcount) {
784 		read_lock(&tasklist_lock);
785 		for_each_process_thread(p, t) {
786 			clear_task_syscall_work(t, SYSCALL_TRACEPOINT);
787 		}
788 		read_unlock(&tasklist_lock);
789 	}
790 }
791 #endif
792