tracing: fix oops in tracepoint_update_probe_range()
[linux-2.6.git] / 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
28 extern struct tracepoint __start___tracepoints[];
29 extern struct tracepoint __stop___tracepoints[];
30
31 /* Set to 1 to enable tracepoint debug output */
32 static const int tracepoint_debug;
33
34 /*
35  * tracepoints_mutex nests inside module_mutex. Tracepoints mutex protects the
36  * builtin and module tracepoints and the hash table.
37  */
38 static DEFINE_MUTEX(tracepoints_mutex);
39
40 /*
41  * Tracepoint hash table, containing the active tracepoints.
42  * Protected by tracepoints_mutex.
43  */
44 #define TRACEPOINT_HASH_BITS 6
45 #define TRACEPOINT_TABLE_SIZE (1 << TRACEPOINT_HASH_BITS)
46 static struct hlist_head tracepoint_table[TRACEPOINT_TABLE_SIZE];
47
48 /*
49  * Note about RCU :
50  * It is used to to delay the free of multiple probes array until a quiescent
51  * state is reached.
52  * Tracepoint entries modifications are protected by the tracepoints_mutex.
53  */
54 struct tracepoint_entry {
55         struct hlist_node hlist;
56         void **funcs;
57         int refcount;   /* Number of times armed. 0 if disarmed. */
58         char name[0];
59 };
60
61 struct tp_probes {
62         union {
63                 struct rcu_head rcu;
64                 struct list_head list;
65         } u;
66         void *probes[0];
67 };
68
69 static inline void *allocate_probes(int count)
70 {
71         struct tp_probes *p  = kmalloc(count * sizeof(void *)
72                         + sizeof(struct tp_probes), GFP_KERNEL);
73         return p == NULL ? NULL : p->probes;
74 }
75
76 static void rcu_free_old_probes(struct rcu_head *head)
77 {
78         kfree(container_of(head, struct tp_probes, u.rcu));
79 }
80
81 static inline void release_probes(void *old)
82 {
83         if (old) {
84                 struct tp_probes *tp_probes = container_of(old,
85                         struct tp_probes, probes[0]);
86                 call_rcu_sched(&tp_probes->u.rcu, rcu_free_old_probes);
87         }
88 }
89
90 static void debug_print_probes(struct tracepoint_entry *entry)
91 {
92         int i;
93
94         if (!tracepoint_debug || !entry->funcs)
95                 return;
96
97         for (i = 0; entry->funcs[i]; i++)
98                 printk(KERN_DEBUG "Probe %d : %p\n", i, entry->funcs[i]);
99 }
100
101 static void *
102 tracepoint_entry_add_probe(struct tracepoint_entry *entry, void *probe)
103 {
104         int nr_probes = 0;
105         void **old, **new;
106
107         WARN_ON(!probe);
108
109         debug_print_probes(entry);
110         old = entry->funcs;
111         if (old) {
112                 /* (N -> N+1), (N != 0, 1) probes */
113                 for (nr_probes = 0; old[nr_probes]; nr_probes++)
114                         if (old[nr_probes] == probe)
115                                 return ERR_PTR(-EEXIST);
116         }
117         /* + 2 : one for new probe, one for NULL func */
118         new = allocate_probes(nr_probes + 2);
119         if (new == NULL)
120                 return ERR_PTR(-ENOMEM);
121         if (old)
122                 memcpy(new, old, nr_probes * sizeof(void *));
123         new[nr_probes] = probe;
124         new[nr_probes + 1] = NULL;
125         entry->refcount = nr_probes + 1;
126         entry->funcs = new;
127         debug_print_probes(entry);
128         return old;
129 }
130
131 static void *
132 tracepoint_entry_remove_probe(struct tracepoint_entry *entry, void *probe)
133 {
134         int nr_probes = 0, nr_del = 0, i;
135         void **old, **new;
136
137         old = entry->funcs;
138
139         if (!old)
140                 return ERR_PTR(-ENOENT);
141
142         debug_print_probes(entry);
143         /* (N -> M), (N > 1, M >= 0) probes */
144         for (nr_probes = 0; old[nr_probes]; nr_probes++) {
145                 if ((!probe || old[nr_probes] == probe))
146                         nr_del++;
147         }
148
149         if (nr_probes - nr_del == 0) {
150                 /* N -> 0, (N > 1) */
151                 entry->funcs = NULL;
152                 entry->refcount = 0;
153                 debug_print_probes(entry);
154                 return old;
155         } else {
156                 int j = 0;
157                 /* N -> M, (N > 1, M > 0) */
158                 /* + 1 for NULL */
159                 new = allocate_probes(nr_probes - nr_del + 1);
160                 if (new == NULL)
161                         return ERR_PTR(-ENOMEM);
162                 for (i = 0; old[i]; i++)
163                         if ((probe && old[i] != probe))
164                                 new[j++] = old[i];
165                 new[nr_probes - nr_del] = NULL;
166                 entry->refcount = nr_probes - nr_del;
167                 entry->funcs = new;
168         }
169         debug_print_probes(entry);
170         return old;
171 }
172
173 /*
174  * Get tracepoint if the tracepoint is present in the tracepoint hash table.
175  * Must be called with tracepoints_mutex held.
176  * Returns NULL if not present.
177  */
178 static struct tracepoint_entry *get_tracepoint(const char *name)
179 {
180         struct hlist_head *head;
181         struct hlist_node *node;
182         struct tracepoint_entry *e;
183         u32 hash = jhash(name, strlen(name), 0);
184
185         head = &tracepoint_table[hash & (TRACEPOINT_TABLE_SIZE - 1)];
186         hlist_for_each_entry(e, node, head, hlist) {
187                 if (!strcmp(name, e->name))
188                         return e;
189         }
190         return NULL;
191 }
192
193 /*
194  * Add the tracepoint to the tracepoint hash table. Must be called with
195  * tracepoints_mutex held.
196  */
197 static struct tracepoint_entry *add_tracepoint(const char *name)
198 {
199         struct hlist_head *head;
200         struct hlist_node *node;
201         struct tracepoint_entry *e;
202         size_t name_len = strlen(name) + 1;
203         u32 hash = jhash(name, name_len-1, 0);
204
205         head = &tracepoint_table[hash & (TRACEPOINT_TABLE_SIZE - 1)];
206         hlist_for_each_entry(e, node, head, hlist) {
207                 if (!strcmp(name, e->name)) {
208                         printk(KERN_NOTICE
209                                 "tracepoint %s busy\n", name);
210                         return ERR_PTR(-EEXIST);        /* Already there */
211                 }
212         }
213         /*
214          * Using kmalloc here to allocate a variable length element. Could
215          * cause some memory fragmentation if overused.
216          */
217         e = kmalloc(sizeof(struct tracepoint_entry) + name_len, GFP_KERNEL);
218         if (!e)
219                 return ERR_PTR(-ENOMEM);
220         memcpy(&e->name[0], name, name_len);
221         e->funcs = NULL;
222         e->refcount = 0;
223         hlist_add_head(&e->hlist, head);
224         return e;
225 }
226
227 /*
228  * Remove the tracepoint from the tracepoint hash table. Must be called with
229  * mutex_lock held.
230  */
231 static inline void remove_tracepoint(struct tracepoint_entry *e)
232 {
233         hlist_del(&e->hlist);
234         kfree(e);
235 }
236
237 /*
238  * Sets the probe callback corresponding to one tracepoint.
239  */
240 static void set_tracepoint(struct tracepoint_entry **entry,
241         struct tracepoint *elem, int active)
242 {
243         WARN_ON(strcmp((*entry)->name, elem->name) != 0);
244
245         /*
246          * rcu_assign_pointer has a smp_wmb() which makes sure that the new
247          * probe callbacks array is consistent before setting a pointer to it.
248          * This array is referenced by __DO_TRACE from
249          * include/linux/tracepoints.h. A matching smp_read_barrier_depends()
250          * is used.
251          */
252         rcu_assign_pointer(elem->funcs, (*entry)->funcs);
253         elem->state = active;
254 }
255
256 /*
257  * Disable a tracepoint and its probe callback.
258  * Note: only waiting an RCU period after setting elem->call to the empty
259  * function insures that the original callback is not used anymore. This insured
260  * by preempt_disable around the call site.
261  */
262 static void disable_tracepoint(struct tracepoint *elem)
263 {
264         elem->state = 0;
265         rcu_assign_pointer(elem->funcs, NULL);
266 }
267
268 /**
269  * tracepoint_update_probe_range - Update a probe range
270  * @begin: beginning of the range
271  * @end: end of the range
272  *
273  * Updates the probe callback corresponding to a range of tracepoints.
274  */
275 void
276 tracepoint_update_probe_range(struct tracepoint *begin, struct tracepoint *end)
277 {
278         struct tracepoint *iter;
279         struct tracepoint_entry *mark_entry;
280
281         if (!begin) {
282                 WARN_ON_ONCE(1);
283                 return;
284         }
285
286         mutex_lock(&tracepoints_mutex);
287         for (iter = begin; iter < end; iter++) {
288                 mark_entry = get_tracepoint(iter->name);
289                 if (mark_entry) {
290                         set_tracepoint(&mark_entry, iter,
291                                         !!mark_entry->refcount);
292                 } else {
293                         disable_tracepoint(iter);
294                 }
295         }
296         mutex_unlock(&tracepoints_mutex);
297 }
298
299 /*
300  * Update probes, removing the faulty probes.
301  */
302 static void tracepoint_update_probes(void)
303 {
304         /* Core kernel tracepoints */
305         tracepoint_update_probe_range(__start___tracepoints,
306                 __stop___tracepoints);
307         /* tracepoints in modules. */
308         module_update_tracepoints();
309 }
310
311 static void *tracepoint_add_probe(const char *name, void *probe)
312 {
313         struct tracepoint_entry *entry;
314         void *old;
315
316         entry = get_tracepoint(name);
317         if (!entry) {
318                 entry = add_tracepoint(name);
319                 if (IS_ERR(entry))
320                         return entry;
321         }
322         old = tracepoint_entry_add_probe(entry, probe);
323         if (IS_ERR(old) && !entry->refcount)
324                 remove_tracepoint(entry);
325         return old;
326 }
327
328 /**
329  * tracepoint_probe_register -  Connect a probe to a tracepoint
330  * @name: tracepoint name
331  * @probe: probe handler
332  *
333  * Returns 0 if ok, error value on error.
334  * The probe address must at least be aligned on the architecture pointer size.
335  */
336 int tracepoint_probe_register(const char *name, void *probe)
337 {
338         void *old;
339
340         mutex_lock(&tracepoints_mutex);
341         old = tracepoint_add_probe(name, probe);
342         mutex_unlock(&tracepoints_mutex);
343         if (IS_ERR(old))
344                 return PTR_ERR(old);
345
346         tracepoint_update_probes();             /* may update entry */
347         release_probes(old);
348         return 0;
349 }
350 EXPORT_SYMBOL_GPL(tracepoint_probe_register);
351
352 static void *tracepoint_remove_probe(const char *name, void *probe)
353 {
354         struct tracepoint_entry *entry;
355         void *old;
356
357         entry = get_tracepoint(name);
358         if (!entry)
359                 return ERR_PTR(-ENOENT);
360         old = tracepoint_entry_remove_probe(entry, probe);
361         if (IS_ERR(old))
362                 return old;
363         if (!entry->refcount)
364                 remove_tracepoint(entry);
365         return old;
366 }
367
368 /**
369  * tracepoint_probe_unregister -  Disconnect a probe from a tracepoint
370  * @name: tracepoint name
371  * @probe: probe function pointer
372  *
373  * We do not need to call a synchronize_sched to make sure the probes have
374  * finished running before doing a module unload, because the module unload
375  * itself uses stop_machine(), which insures that every preempt disabled section
376  * have finished.
377  */
378 int tracepoint_probe_unregister(const char *name, void *probe)
379 {
380         void *old;
381
382         mutex_lock(&tracepoints_mutex);
383         old = tracepoint_remove_probe(name, probe);
384         mutex_unlock(&tracepoints_mutex);
385         if (IS_ERR(old))
386                 return PTR_ERR(old);
387
388         tracepoint_update_probes();             /* may update entry */
389         release_probes(old);
390         return 0;
391 }
392 EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
393
394 static LIST_HEAD(old_probes);
395 static int need_update;
396
397 static void tracepoint_add_old_probes(void *old)
398 {
399         need_update = 1;
400         if (old) {
401                 struct tp_probes *tp_probes = container_of(old,
402                         struct tp_probes, probes[0]);
403                 list_add(&tp_probes->u.list, &old_probes);
404         }
405 }
406
407 /**
408  * tracepoint_probe_register_noupdate -  register a probe but not connect
409  * @name: tracepoint name
410  * @probe: probe handler
411  *
412  * caller must call tracepoint_probe_update_all()
413  */
414 int tracepoint_probe_register_noupdate(const char *name, void *probe)
415 {
416         void *old;
417
418         mutex_lock(&tracepoints_mutex);
419         old = tracepoint_add_probe(name, probe);
420         if (IS_ERR(old)) {
421                 mutex_unlock(&tracepoints_mutex);
422                 return PTR_ERR(old);
423         }
424         tracepoint_add_old_probes(old);
425         mutex_unlock(&tracepoints_mutex);
426         return 0;
427 }
428 EXPORT_SYMBOL_GPL(tracepoint_probe_register_noupdate);
429
430 /**
431  * tracepoint_probe_unregister_noupdate -  remove a probe but not disconnect
432  * @name: tracepoint name
433  * @probe: probe function pointer
434  *
435  * caller must call tracepoint_probe_update_all()
436  */
437 int tracepoint_probe_unregister_noupdate(const char *name, void *probe)
438 {
439         void *old;
440
441         mutex_lock(&tracepoints_mutex);
442         old = tracepoint_remove_probe(name, probe);
443         if (IS_ERR(old)) {
444                 mutex_unlock(&tracepoints_mutex);
445                 return PTR_ERR(old);
446         }
447         tracepoint_add_old_probes(old);
448         mutex_unlock(&tracepoints_mutex);
449         return 0;
450 }
451 EXPORT_SYMBOL_GPL(tracepoint_probe_unregister_noupdate);
452
453 /**
454  * tracepoint_probe_update_all -  update tracepoints
455  */
456 void tracepoint_probe_update_all(void)
457 {
458         LIST_HEAD(release_probes);
459         struct tp_probes *pos, *next;
460
461         mutex_lock(&tracepoints_mutex);
462         if (!need_update) {
463                 mutex_unlock(&tracepoints_mutex);
464                 return;
465         }
466         if (!list_empty(&old_probes))
467                 list_replace_init(&old_probes, &release_probes);
468         need_update = 0;
469         mutex_unlock(&tracepoints_mutex);
470
471         tracepoint_update_probes();
472         list_for_each_entry_safe(pos, next, &release_probes, u.list) {
473                 list_del(&pos->u.list);
474                 call_rcu_sched(&pos->u.rcu, rcu_free_old_probes);
475         }
476 }
477 EXPORT_SYMBOL_GPL(tracepoint_probe_update_all);
478
479 /**
480  * tracepoint_get_iter_range - Get a next tracepoint iterator given a range.
481  * @tracepoint: current tracepoints (in), next tracepoint (out)
482  * @begin: beginning of the range
483  * @end: end of the range
484  *
485  * Returns whether a next tracepoint has been found (1) or not (0).
486  * Will return the first tracepoint in the range if the input tracepoint is
487  * NULL.
488  */
489 int tracepoint_get_iter_range(struct tracepoint **tracepoint,
490         struct tracepoint *begin, struct tracepoint *end)
491 {
492         if (!*tracepoint && begin != end) {
493                 *tracepoint = begin;
494                 return 1;
495         }
496         if (*tracepoint >= begin && *tracepoint < end)
497                 return 1;
498         return 0;
499 }
500 EXPORT_SYMBOL_GPL(tracepoint_get_iter_range);
501
502 static void tracepoint_get_iter(struct tracepoint_iter *iter)
503 {
504         int found = 0;
505
506         /* Core kernel tracepoints */
507         if (!iter->module) {
508                 found = tracepoint_get_iter_range(&iter->tracepoint,
509                                 __start___tracepoints, __stop___tracepoints);
510                 if (found)
511                         goto end;
512         }
513         /* tracepoints in modules. */
514         found = module_get_iter_tracepoints(iter);
515 end:
516         if (!found)
517                 tracepoint_iter_reset(iter);
518 }
519
520 void tracepoint_iter_start(struct tracepoint_iter *iter)
521 {
522         tracepoint_get_iter(iter);
523 }
524 EXPORT_SYMBOL_GPL(tracepoint_iter_start);
525
526 void tracepoint_iter_next(struct tracepoint_iter *iter)
527 {
528         iter->tracepoint++;
529         /*
530          * iter->tracepoint may be invalid because we blindly incremented it.
531          * Make sure it is valid by marshalling on the tracepoints, getting the
532          * tracepoints from following modules if necessary.
533          */
534         tracepoint_get_iter(iter);
535 }
536 EXPORT_SYMBOL_GPL(tracepoint_iter_next);
537
538 void tracepoint_iter_stop(struct tracepoint_iter *iter)
539 {
540 }
541 EXPORT_SYMBOL_GPL(tracepoint_iter_stop);
542
543 void tracepoint_iter_reset(struct tracepoint_iter *iter)
544 {
545         iter->module = NULL;
546         iter->tracepoint = NULL;
547 }
548 EXPORT_SYMBOL_GPL(tracepoint_iter_reset);
549
550 #ifdef CONFIG_MODULES
551
552 int tracepoint_module_notify(struct notifier_block *self,
553                              unsigned long val, void *data)
554 {
555         struct module *mod = data;
556
557         switch (val) {
558         case MODULE_STATE_COMING:
559                 tracepoint_update_probe_range(mod->tracepoints,
560                         mod->tracepoints + mod->num_tracepoints);
561                 break;
562         case MODULE_STATE_GOING:
563                 tracepoint_update_probe_range(mod->tracepoints,
564                         mod->tracepoints + mod->num_tracepoints);
565                 break;
566         }
567         return 0;
568 }
569
570 struct notifier_block tracepoint_module_nb = {
571         .notifier_call = tracepoint_module_notify,
572         .priority = 0,
573 };
574
575 static int init_tracepoints(void)
576 {
577         return register_module_notifier(&tracepoint_module_nb);
578 }
579 __initcall(init_tracepoints);
580
581 #endif /* CONFIG_MODULES */