blob: 7d1faecd7a51bd643cc1e10f0f2048765136bf16 [file] [log] [blame]
Mathieu Desnoyers8256e472007-10-18 23:41:06 -07001/*
2 * Copyright (C) 2007 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/marker.h>
25#include <linux/err.h>
Robert P. J. Day1aeb2722008-04-29 00:59:25 -070026#include <linux/slab.h>
Mathieu Desnoyers8256e472007-10-18 23:41:06 -070027
28extern struct marker __start___markers[];
29extern struct marker __stop___markers[];
30
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -080031/* Set to 1 to enable marker debug output */
Adrian Bunkab883af2008-04-30 00:54:30 -070032static const int marker_debug;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -070033
34/*
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -080035 * markers_mutex nests inside module_mutex. Markers mutex protects the builtin
36 * and module markers and the hash table.
Mathieu Desnoyers8256e472007-10-18 23:41:06 -070037 */
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -080038static DEFINE_MUTEX(markers_mutex);
Mathieu Desnoyers8256e472007-10-18 23:41:06 -070039
40/*
41 * Marker hash table, containing the active markers.
42 * Protected by module_mutex.
43 */
44#define MARKER_HASH_BITS 6
45#define MARKER_TABLE_SIZE (1 << MARKER_HASH_BITS)
46
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -080047/*
48 * Note about RCU :
49 * It is used to make sure every handler has finished using its private data
50 * between two consecutive operation (add or remove) on a given marker. It is
51 * also used to delay the free of multiple probes array until a quiescent state
52 * is reached.
53 * marker entries modifications are protected by the markers_mutex.
54 */
Mathieu Desnoyers8256e472007-10-18 23:41:06 -070055struct marker_entry {
56 struct hlist_node hlist;
57 char *format;
Mathieu Desnoyersdc102a82008-05-12 21:21:09 +020058 /* Probe wrapper */
59 void (*call)(const struct marker *mdata, void *call_private, ...);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -080060 struct marker_probe_closure single;
61 struct marker_probe_closure *multi;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -070062 int refcount; /* Number of times armed. 0 if disarmed. */
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -080063 struct rcu_head rcu;
64 void *oldptr;
Harvey Harrisonde4fc642008-02-23 15:23:33 -080065 unsigned char rcu_pending:1;
66 unsigned char ptype:1;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -070067 char name[0]; /* Contains name'\0'format'\0' */
68};
69
70static struct hlist_head marker_table[MARKER_TABLE_SIZE];
71
72/**
73 * __mark_empty_function - Empty probe callback
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -080074 * @probe_private: probe private data
75 * @call_private: call site private data
Mathieu Desnoyers8256e472007-10-18 23:41:06 -070076 * @fmt: format string
77 * @...: variable argument list
78 *
79 * Empty callback provided as a probe to the markers. By providing this to a
80 * disabled marker, we make sure the execution flow is always valid even
81 * though the function pointer change and the marker enabling are two distinct
82 * operations that modifies the execution flow of preemptible code.
83 */
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -080084void __mark_empty_function(void *probe_private, void *call_private,
85 const char *fmt, va_list *args)
Mathieu Desnoyers8256e472007-10-18 23:41:06 -070086{
87}
88EXPORT_SYMBOL_GPL(__mark_empty_function);
89
90/*
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -080091 * marker_probe_cb Callback that prepares the variable argument list for probes.
92 * @mdata: pointer of type struct marker
93 * @call_private: caller site private data
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -080094 * @...: Variable argument list.
95 *
96 * Since we do not use "typical" pointer based RCU in the 1 argument case, we
97 * need to put a full smp_rmb() in this branch. This is why we do not use
98 * rcu_dereference() for the pointer read.
99 */
Mathieu Desnoyersdc102a82008-05-12 21:21:09 +0200100void marker_probe_cb(const struct marker *mdata, void *call_private, ...)
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800101{
102 va_list args;
103 char ptype;
104
105 /*
Mathieu Desnoyersfd3c36f2008-03-24 12:29:47 -0700106 * preempt_disable does two things : disabling preemption to make sure
107 * the teardown of the callbacks can be done correctly when they are in
108 * modules and they insure RCU read coherency.
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800109 */
110 preempt_disable();
Mathieu Desnoyers58336112008-03-24 12:29:49 -0700111 ptype = mdata->ptype;
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800112 if (likely(!ptype)) {
113 marker_probe_func *func;
114 /* Must read the ptype before ptr. They are not data dependant,
115 * so we put an explicit smp_rmb() here. */
116 smp_rmb();
Mathieu Desnoyers58336112008-03-24 12:29:49 -0700117 func = mdata->single.func;
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800118 /* Must read the ptr before private data. They are not data
119 * dependant, so we put an explicit smp_rmb() here. */
120 smp_rmb();
Mathieu Desnoyersdc102a82008-05-12 21:21:09 +0200121 va_start(args, call_private);
122 func(mdata->single.probe_private, call_private, mdata->format,
123 &args);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800124 va_end(args);
125 } else {
126 struct marker_probe_closure *multi;
127 int i;
128 /*
Mathieu Desnoyers5def9a32008-07-29 22:33:31 -0700129 * Read mdata->ptype before mdata->multi.
130 */
131 smp_rmb();
132 multi = mdata->multi;
133 /*
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800134 * multi points to an array, therefore accessing the array
135 * depends on reading multi. However, even in this case,
136 * we must insure that the pointer is read _before_ the array
137 * data. Same as rcu_dereference, but we need a full smp_rmb()
138 * in the fast path, so put the explicit barrier here.
139 */
140 smp_read_barrier_depends();
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800141 for (i = 0; multi[i].func; i++) {
Mathieu Desnoyersdc102a82008-05-12 21:21:09 +0200142 va_start(args, call_private);
143 multi[i].func(multi[i].probe_private, call_private,
144 mdata->format, &args);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800145 va_end(args);
146 }
147 }
148 preempt_enable();
149}
150EXPORT_SYMBOL_GPL(marker_probe_cb);
151
152/*
153 * marker_probe_cb Callback that does not prepare the variable argument list.
154 * @mdata: pointer of type struct marker
155 * @call_private: caller site private data
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800156 * @...: Variable argument list.
157 *
158 * Should be connected to markers "MARK_NOARGS".
159 */
Mathieu Desnoyersdc102a82008-05-12 21:21:09 +0200160void marker_probe_cb_noarg(const struct marker *mdata, void *call_private, ...)
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800161{
162 va_list args; /* not initialized */
163 char ptype;
164
165 preempt_disable();
Mathieu Desnoyers58336112008-03-24 12:29:49 -0700166 ptype = mdata->ptype;
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800167 if (likely(!ptype)) {
168 marker_probe_func *func;
169 /* Must read the ptype before ptr. They are not data dependant,
170 * so we put an explicit smp_rmb() here. */
171 smp_rmb();
Mathieu Desnoyers58336112008-03-24 12:29:49 -0700172 func = mdata->single.func;
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800173 /* Must read the ptr before private data. They are not data
174 * dependant, so we put an explicit smp_rmb() here. */
175 smp_rmb();
Mathieu Desnoyersdc102a82008-05-12 21:21:09 +0200176 func(mdata->single.probe_private, call_private, mdata->format,
177 &args);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800178 } else {
179 struct marker_probe_closure *multi;
180 int i;
181 /*
Mathieu Desnoyers5def9a32008-07-29 22:33:31 -0700182 * Read mdata->ptype before mdata->multi.
183 */
184 smp_rmb();
185 multi = mdata->multi;
186 /*
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800187 * multi points to an array, therefore accessing the array
188 * depends on reading multi. However, even in this case,
189 * we must insure that the pointer is read _before_ the array
190 * data. Same as rcu_dereference, but we need a full smp_rmb()
191 * in the fast path, so put the explicit barrier here.
192 */
193 smp_read_barrier_depends();
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800194 for (i = 0; multi[i].func; i++)
Mathieu Desnoyersdc102a82008-05-12 21:21:09 +0200195 multi[i].func(multi[i].probe_private, call_private,
196 mdata->format, &args);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800197 }
198 preempt_enable();
199}
200EXPORT_SYMBOL_GPL(marker_probe_cb_noarg);
201
202static void free_old_closure(struct rcu_head *head)
203{
204 struct marker_entry *entry = container_of(head,
205 struct marker_entry, rcu);
206 kfree(entry->oldptr);
207 /* Make sure we free the data before setting the pending flag to 0 */
208 smp_wmb();
209 entry->rcu_pending = 0;
210}
211
212static void debug_print_probes(struct marker_entry *entry)
213{
214 int i;
215
216 if (!marker_debug)
217 return;
218
219 if (!entry->ptype) {
220 printk(KERN_DEBUG "Single probe : %p %p\n",
221 entry->single.func,
222 entry->single.probe_private);
223 } else {
224 for (i = 0; entry->multi[i].func; i++)
225 printk(KERN_DEBUG "Multi probe %d : %p %p\n", i,
226 entry->multi[i].func,
227 entry->multi[i].probe_private);
228 }
229}
230
231static struct marker_probe_closure *
232marker_entry_add_probe(struct marker_entry *entry,
233 marker_probe_func *probe, void *probe_private)
234{
235 int nr_probes = 0;
236 struct marker_probe_closure *old, *new;
237
238 WARN_ON(!probe);
239
240 debug_print_probes(entry);
241 old = entry->multi;
242 if (!entry->ptype) {
243 if (entry->single.func == probe &&
244 entry->single.probe_private == probe_private)
245 return ERR_PTR(-EBUSY);
246 if (entry->single.func == __mark_empty_function) {
247 /* 0 -> 1 probes */
248 entry->single.func = probe;
249 entry->single.probe_private = probe_private;
250 entry->refcount = 1;
251 entry->ptype = 0;
252 debug_print_probes(entry);
253 return NULL;
254 } else {
255 /* 1 -> 2 probes */
256 nr_probes = 1;
257 old = NULL;
258 }
259 } else {
260 /* (N -> N+1), (N != 0, 1) probes */
261 for (nr_probes = 0; old[nr_probes].func; nr_probes++)
262 if (old[nr_probes].func == probe
263 && old[nr_probes].probe_private
264 == probe_private)
265 return ERR_PTR(-EBUSY);
266 }
267 /* + 2 : one for new probe, one for NULL func */
268 new = kzalloc((nr_probes + 2) * sizeof(struct marker_probe_closure),
269 GFP_KERNEL);
270 if (new == NULL)
271 return ERR_PTR(-ENOMEM);
272 if (!old)
273 new[0] = entry->single;
274 else
275 memcpy(new, old,
276 nr_probes * sizeof(struct marker_probe_closure));
277 new[nr_probes].func = probe;
278 new[nr_probes].probe_private = probe_private;
279 entry->refcount = nr_probes + 1;
280 entry->multi = new;
281 entry->ptype = 1;
282 debug_print_probes(entry);
283 return old;
284}
285
286static struct marker_probe_closure *
287marker_entry_remove_probe(struct marker_entry *entry,
288 marker_probe_func *probe, void *probe_private)
289{
290 int nr_probes = 0, nr_del = 0, i;
291 struct marker_probe_closure *old, *new;
292
293 old = entry->multi;
294
295 debug_print_probes(entry);
296 if (!entry->ptype) {
297 /* 0 -> N is an error */
298 WARN_ON(entry->single.func == __mark_empty_function);
299 /* 1 -> 0 probes */
300 WARN_ON(probe && entry->single.func != probe);
301 WARN_ON(entry->single.probe_private != probe_private);
302 entry->single.func = __mark_empty_function;
303 entry->refcount = 0;
304 entry->ptype = 0;
305 debug_print_probes(entry);
306 return NULL;
307 } else {
308 /* (N -> M), (N > 1, M >= 0) probes */
309 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
310 if ((!probe || old[nr_probes].func == probe)
311 && old[nr_probes].probe_private
312 == probe_private)
313 nr_del++;
314 }
315 }
316
317 if (nr_probes - nr_del == 0) {
318 /* N -> 0, (N > 1) */
319 entry->single.func = __mark_empty_function;
320 entry->refcount = 0;
321 entry->ptype = 0;
322 } else if (nr_probes - nr_del == 1) {
323 /* N -> 1, (N > 1) */
324 for (i = 0; old[i].func; i++)
325 if ((probe && old[i].func != probe) ||
326 old[i].probe_private != probe_private)
327 entry->single = old[i];
328 entry->refcount = 1;
329 entry->ptype = 0;
330 } else {
331 int j = 0;
332 /* N -> M, (N > 1, M > 1) */
333 /* + 1 for NULL */
334 new = kzalloc((nr_probes - nr_del + 1)
335 * sizeof(struct marker_probe_closure), GFP_KERNEL);
336 if (new == NULL)
337 return ERR_PTR(-ENOMEM);
338 for (i = 0; old[i].func; i++)
339 if ((probe && old[i].func != probe) ||
340 old[i].probe_private != probe_private)
341 new[j++] = old[i];
342 entry->refcount = nr_probes - nr_del;
343 entry->ptype = 1;
344 entry->multi = new;
345 }
346 debug_print_probes(entry);
347 return old;
348}
349
350/*
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700351 * Get marker if the marker is present in the marker hash table.
352 * Must be called with markers_mutex held.
353 * Returns NULL if not present.
354 */
355static struct marker_entry *get_marker(const char *name)
356{
357 struct hlist_head *head;
358 struct hlist_node *node;
359 struct marker_entry *e;
360 u32 hash = jhash(name, strlen(name), 0);
361
362 head = &marker_table[hash & ((1 << MARKER_HASH_BITS)-1)];
363 hlist_for_each_entry(e, node, head, hlist) {
364 if (!strcmp(name, e->name))
365 return e;
366 }
367 return NULL;
368}
369
370/*
371 * Add the marker to the marker hash table. Must be called with markers_mutex
372 * held.
373 */
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800374static struct marker_entry *add_marker(const char *name, const char *format)
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700375{
376 struct hlist_head *head;
377 struct hlist_node *node;
378 struct marker_entry *e;
379 size_t name_len = strlen(name) + 1;
380 size_t format_len = 0;
381 u32 hash = jhash(name, name_len-1, 0);
382
383 if (format)
384 format_len = strlen(format) + 1;
385 head = &marker_table[hash & ((1 << MARKER_HASH_BITS)-1)];
386 hlist_for_each_entry(e, node, head, hlist) {
387 if (!strcmp(name, e->name)) {
388 printk(KERN_NOTICE
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800389 "Marker %s busy\n", name);
390 return ERR_PTR(-EBUSY); /* Already there */
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700391 }
392 }
393 /*
394 * Using kmalloc here to allocate a variable length element. Could
395 * cause some memory fragmentation if overused.
396 */
397 e = kmalloc(sizeof(struct marker_entry) + name_len + format_len,
398 GFP_KERNEL);
399 if (!e)
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800400 return ERR_PTR(-ENOMEM);
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700401 memcpy(&e->name[0], name, name_len);
402 if (format) {
403 e->format = &e->name[name_len];
404 memcpy(e->format, format, format_len);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800405 if (strcmp(e->format, MARK_NOARGS) == 0)
406 e->call = marker_probe_cb_noarg;
407 else
408 e->call = marker_probe_cb;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700409 trace_mark(core_marker_format, "name %s format %s",
410 e->name, e->format);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800411 } else {
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700412 e->format = NULL;
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800413 e->call = marker_probe_cb;
414 }
415 e->single.func = __mark_empty_function;
416 e->single.probe_private = NULL;
417 e->multi = NULL;
418 e->ptype = 0;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700419 e->refcount = 0;
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800420 e->rcu_pending = 0;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700421 hlist_add_head(&e->hlist, head);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800422 return e;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700423}
424
425/*
426 * Remove the marker from the marker hash table. Must be called with mutex_lock
427 * held.
428 */
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800429static int remove_marker(const char *name)
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700430{
431 struct hlist_head *head;
432 struct hlist_node *node;
433 struct marker_entry *e;
434 int found = 0;
435 size_t len = strlen(name) + 1;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700436 u32 hash = jhash(name, len-1, 0);
437
438 head = &marker_table[hash & ((1 << MARKER_HASH_BITS)-1)];
439 hlist_for_each_entry(e, node, head, hlist) {
440 if (!strcmp(name, e->name)) {
441 found = 1;
442 break;
443 }
444 }
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800445 if (!found)
446 return -ENOENT;
447 if (e->single.func != __mark_empty_function)
448 return -EBUSY;
449 hlist_del(&e->hlist);
450 /* Make sure the call_rcu has been executed */
451 if (e->rcu_pending)
Mathieu Desnoyers28325df2008-07-25 01:48:38 -0700452 rcu_barrier_sched();
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800453 kfree(e);
454 return 0;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700455}
456
457/*
458 * Set the mark_entry format to the format found in the element.
459 */
460static int marker_set_format(struct marker_entry **entry, const char *format)
461{
462 struct marker_entry *e;
463 size_t name_len = strlen((*entry)->name) + 1;
464 size_t format_len = strlen(format) + 1;
465
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800466
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700467 e = kmalloc(sizeof(struct marker_entry) + name_len + format_len,
468 GFP_KERNEL);
469 if (!e)
470 return -ENOMEM;
471 memcpy(&e->name[0], (*entry)->name, name_len);
472 e->format = &e->name[name_len];
473 memcpy(e->format, format, format_len);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800474 if (strcmp(e->format, MARK_NOARGS) == 0)
475 e->call = marker_probe_cb_noarg;
476 else
477 e->call = marker_probe_cb;
478 e->single = (*entry)->single;
479 e->multi = (*entry)->multi;
480 e->ptype = (*entry)->ptype;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700481 e->refcount = (*entry)->refcount;
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800482 e->rcu_pending = 0;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700483 hlist_add_before(&e->hlist, &(*entry)->hlist);
484 hlist_del(&(*entry)->hlist);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800485 /* Make sure the call_rcu has been executed */
486 if ((*entry)->rcu_pending)
Mathieu Desnoyers28325df2008-07-25 01:48:38 -0700487 rcu_barrier_sched();
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700488 kfree(*entry);
489 *entry = e;
490 trace_mark(core_marker_format, "name %s format %s",
491 e->name, e->format);
492 return 0;
493}
494
495/*
496 * Sets the probe callback corresponding to one marker.
497 */
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800498static int set_marker(struct marker_entry **entry, struct marker *elem,
499 int active)
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700500{
501 int ret;
502 WARN_ON(strcmp((*entry)->name, elem->name) != 0);
503
504 if ((*entry)->format) {
505 if (strcmp((*entry)->format, elem->format) != 0) {
506 printk(KERN_NOTICE
507 "Format mismatch for probe %s "
508 "(%s), marker (%s)\n",
509 (*entry)->name,
510 (*entry)->format,
511 elem->format);
512 return -EPERM;
513 }
514 } else {
515 ret = marker_set_format(entry, elem->format);
516 if (ret)
517 return ret;
518 }
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800519
520 /*
521 * probe_cb setup (statically known) is done here. It is
522 * asynchronous with the rest of execution, therefore we only
523 * pass from a "safe" callback (with argument) to an "unsafe"
524 * callback (does not set arguments).
525 */
526 elem->call = (*entry)->call;
527 /*
528 * Sanity check :
529 * We only update the single probe private data when the ptr is
530 * set to a _non_ single probe! (0 -> 1 and N -> 1, N != 1)
531 */
532 WARN_ON(elem->single.func != __mark_empty_function
533 && elem->single.probe_private
534 != (*entry)->single.probe_private &&
535 !elem->ptype);
536 elem->single.probe_private = (*entry)->single.probe_private;
537 /*
538 * Make sure the private data is valid when we update the
539 * single probe ptr.
540 */
541 smp_wmb();
542 elem->single.func = (*entry)->single.func;
543 /*
544 * We also make sure that the new probe callbacks array is consistent
545 * before setting a pointer to it.
546 */
547 rcu_assign_pointer(elem->multi, (*entry)->multi);
548 /*
549 * Update the function or multi probe array pointer before setting the
550 * ptype.
551 */
552 smp_wmb();
553 elem->ptype = (*entry)->ptype;
554 elem->state = active;
555
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700556 return 0;
557}
558
559/*
560 * Disable a marker and its probe callback.
Mathieu Desnoyersfd3c36f2008-03-24 12:29:47 -0700561 * Note: only waiting an RCU period after setting elem->call to the empty
562 * function insures that the original callback is not used anymore. This insured
563 * by preempt_disable around the call site.
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700564 */
565static void disable_marker(struct marker *elem)
566{
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800567 /* leave "call" as is. It is known statically. */
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700568 elem->state = 0;
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800569 elem->single.func = __mark_empty_function;
570 /* Update the function before setting the ptype */
571 smp_wmb();
572 elem->ptype = 0; /* single probe */
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700573 /*
574 * Leave the private data and id there, because removal is racy and
Mathieu Desnoyersfd3c36f2008-03-24 12:29:47 -0700575 * should be done only after an RCU period. These are never used until
576 * the next initialization anyway.
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700577 */
578}
579
580/**
581 * marker_update_probe_range - Update a probe range
582 * @begin: beginning of the range
583 * @end: end of the range
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700584 *
585 * Updates the probe callback corresponding to a range of markers.
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700586 */
587void marker_update_probe_range(struct marker *begin,
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800588 struct marker *end)
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700589{
590 struct marker *iter;
591 struct marker_entry *mark_entry;
592
Mathieu Desnoyers314de8a2007-11-14 16:59:48 -0800593 mutex_lock(&markers_mutex);
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700594 for (iter = begin; iter < end; iter++) {
595 mark_entry = get_marker(iter->name);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800596 if (mark_entry) {
597 set_marker(&mark_entry, iter,
598 !!mark_entry->refcount);
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700599 /*
600 * ignore error, continue
601 */
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700602 } else {
603 disable_marker(iter);
604 }
605 }
Mathieu Desnoyers314de8a2007-11-14 16:59:48 -0800606 mutex_unlock(&markers_mutex);
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700607}
608
609/*
610 * Update probes, removing the faulty probes.
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800611 *
612 * Internal callback only changed before the first probe is connected to it.
613 * Single probe private data can only be changed on 0 -> 1 and 2 -> 1
614 * transitions. All other transitions will leave the old private data valid.
615 * This makes the non-atomicity of the callback/private data updates valid.
616 *
617 * "special case" updates :
618 * 0 -> 1 callback
619 * 1 -> 0 callback
620 * 1 -> 2 callbacks
621 * 2 -> 1 callbacks
622 * Other updates all behave the same, just like the 2 -> 3 or 3 -> 2 updates.
623 * Site effect : marker_set_format may delete the marker entry (creating a
624 * replacement).
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700625 */
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800626static void marker_update_probes(void)
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700627{
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700628 /* Core kernel markers */
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800629 marker_update_probe_range(__start___markers, __stop___markers);
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700630 /* Markers in modules. */
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800631 module_update_markers();
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700632}
633
634/**
635 * marker_probe_register - Connect a probe to a marker
636 * @name: marker name
637 * @format: format string
638 * @probe: probe handler
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800639 * @probe_private: probe private data
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700640 *
641 * private data must be a valid allocated memory address, or NULL.
642 * Returns 0 if ok, error value on error.
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800643 * The probe address must at least be aligned on the architecture pointer size.
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700644 */
645int marker_probe_register(const char *name, const char *format,
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800646 marker_probe_func *probe, void *probe_private)
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700647{
648 struct marker_entry *entry;
Mathieu Desnoyers314de8a2007-11-14 16:59:48 -0800649 int ret = 0;
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800650 struct marker_probe_closure *old;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700651
652 mutex_lock(&markers_mutex);
653 entry = get_marker(name);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800654 if (!entry) {
655 entry = add_marker(name, format);
656 if (IS_ERR(entry)) {
657 ret = PTR_ERR(entry);
658 goto end;
659 }
660 }
661 /*
662 * If we detect that a call_rcu is pending for this marker,
663 * make sure it's executed now.
664 */
665 if (entry->rcu_pending)
Mathieu Desnoyers28325df2008-07-25 01:48:38 -0700666 rcu_barrier_sched();
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800667 old = marker_entry_add_probe(entry, probe, probe_private);
668 if (IS_ERR(old)) {
669 ret = PTR_ERR(old);
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700670 goto end;
671 }
Mathieu Desnoyers314de8a2007-11-14 16:59:48 -0800672 mutex_unlock(&markers_mutex);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800673 marker_update_probes(); /* may update entry */
674 mutex_lock(&markers_mutex);
675 entry = get_marker(name);
676 WARN_ON(!entry);
677 entry->oldptr = old;
678 entry->rcu_pending = 1;
679 /* write rcu_pending before calling the RCU callback */
680 smp_wmb();
Mathieu Desnoyers28325df2008-07-25 01:48:38 -0700681 call_rcu_sched(&entry->rcu, free_old_closure);
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700682end:
683 mutex_unlock(&markers_mutex);
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700684 return ret;
685}
686EXPORT_SYMBOL_GPL(marker_probe_register);
687
688/**
689 * marker_probe_unregister - Disconnect a probe from a marker
690 * @name: marker name
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800691 * @probe: probe function pointer
692 * @probe_private: probe private data
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700693 *
694 * Returns the private data given to marker_probe_register, or an ERR_PTR().
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800695 * We do not need to call a synchronize_sched to make sure the probes have
696 * finished running before doing a module unload, because the module unload
697 * itself uses stop_machine(), which insures that every preempt disabled section
698 * have finished.
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700699 */
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800700int marker_probe_unregister(const char *name,
701 marker_probe_func *probe, void *probe_private)
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700702{
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700703 struct marker_entry *entry;
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800704 struct marker_probe_closure *old;
Jesper Juhl544adb42008-03-04 14:29:00 -0800705 int ret = -ENOENT;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700706
707 mutex_lock(&markers_mutex);
708 entry = get_marker(name);
Jesper Juhl544adb42008-03-04 14:29:00 -0800709 if (!entry)
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700710 goto end;
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800711 if (entry->rcu_pending)
Mathieu Desnoyers28325df2008-07-25 01:48:38 -0700712 rcu_barrier_sched();
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800713 old = marker_entry_remove_probe(entry, probe, probe_private);
Mathieu Desnoyers314de8a2007-11-14 16:59:48 -0800714 mutex_unlock(&markers_mutex);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800715 marker_update_probes(); /* may update entry */
716 mutex_lock(&markers_mutex);
717 entry = get_marker(name);
Jesper Juhl544adb42008-03-04 14:29:00 -0800718 if (!entry)
719 goto end;
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800720 entry->oldptr = old;
721 entry->rcu_pending = 1;
722 /* write rcu_pending before calling the RCU callback */
723 smp_wmb();
Mathieu Desnoyers28325df2008-07-25 01:48:38 -0700724 call_rcu_sched(&entry->rcu, free_old_closure);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800725 remove_marker(name); /* Ignore busy error message */
Jesper Juhl544adb42008-03-04 14:29:00 -0800726 ret = 0;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700727end:
728 mutex_unlock(&markers_mutex);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800729 return ret;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700730}
731EXPORT_SYMBOL_GPL(marker_probe_unregister);
732
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800733static struct marker_entry *
734get_marker_from_private_data(marker_probe_func *probe, void *probe_private)
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700735{
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800736 struct marker_entry *entry;
737 unsigned int i;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700738 struct hlist_head *head;
739 struct hlist_node *node;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700740
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700741 for (i = 0; i < MARKER_TABLE_SIZE; i++) {
742 head = &marker_table[i];
743 hlist_for_each_entry(entry, node, head, hlist) {
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800744 if (!entry->ptype) {
745 if (entry->single.func == probe
746 && entry->single.probe_private
747 == probe_private)
748 return entry;
749 } else {
750 struct marker_probe_closure *closure;
751 closure = entry->multi;
752 for (i = 0; closure[i].func; i++) {
753 if (closure[i].func == probe &&
754 closure[i].probe_private
755 == probe_private)
756 return entry;
757 }
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700758 }
759 }
760 }
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800761 return NULL;
762}
763
764/**
765 * marker_probe_unregister_private_data - Disconnect a probe from a marker
766 * @probe: probe function
767 * @probe_private: probe private data
768 *
769 * Unregister a probe by providing the registered private data.
770 * Only removes the first marker found in hash table.
771 * Return 0 on success or error value.
772 * We do not need to call a synchronize_sched to make sure the probes have
773 * finished running before doing a module unload, because the module unload
774 * itself uses stop_machine(), which insures that every preempt disabled section
775 * have finished.
776 */
777int marker_probe_unregister_private_data(marker_probe_func *probe,
778 void *probe_private)
779{
780 struct marker_entry *entry;
781 int ret = 0;
782 struct marker_probe_closure *old;
783
784 mutex_lock(&markers_mutex);
785 entry = get_marker_from_private_data(probe, probe_private);
786 if (!entry) {
787 ret = -ENOENT;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700788 goto end;
789 }
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800790 if (entry->rcu_pending)
Mathieu Desnoyers28325df2008-07-25 01:48:38 -0700791 rcu_barrier_sched();
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800792 old = marker_entry_remove_probe(entry, NULL, probe_private);
Mathieu Desnoyers314de8a2007-11-14 16:59:48 -0800793 mutex_unlock(&markers_mutex);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800794 marker_update_probes(); /* may update entry */
795 mutex_lock(&markers_mutex);
796 entry = get_marker_from_private_data(probe, probe_private);
797 WARN_ON(!entry);
798 entry->oldptr = old;
799 entry->rcu_pending = 1;
800 /* write rcu_pending before calling the RCU callback */
801 smp_wmb();
Mathieu Desnoyers28325df2008-07-25 01:48:38 -0700802 call_rcu_sched(&entry->rcu, free_old_closure);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800803 remove_marker(entry->name); /* Ignore busy error message */
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700804end:
805 mutex_unlock(&markers_mutex);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800806 return ret;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700807}
808EXPORT_SYMBOL_GPL(marker_probe_unregister_private_data);
809
810/**
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700811 * marker_get_private_data - Get a marker's probe private data
812 * @name: marker name
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800813 * @probe: probe to match
814 * @num: get the nth matching probe's private data
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700815 *
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800816 * Returns the nth private data pointer (starting from 0) matching, or an
817 * ERR_PTR.
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700818 * Returns the private data pointer, or an ERR_PTR.
819 * The private data pointer should _only_ be dereferenced if the caller is the
820 * owner of the data, or its content could vanish. This is mostly used to
821 * confirm that a caller is the owner of a registered probe.
822 */
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800823void *marker_get_private_data(const char *name, marker_probe_func *probe,
824 int num)
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700825{
826 struct hlist_head *head;
827 struct hlist_node *node;
828 struct marker_entry *e;
829 size_t name_len = strlen(name) + 1;
830 u32 hash = jhash(name, name_len-1, 0);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800831 int i;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700832
833 head = &marker_table[hash & ((1 << MARKER_HASH_BITS)-1)];
834 hlist_for_each_entry(e, node, head, hlist) {
835 if (!strcmp(name, e->name)) {
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800836 if (!e->ptype) {
837 if (num == 0 && e->single.func == probe)
838 return e->single.probe_private;
839 else
840 break;
841 } else {
842 struct marker_probe_closure *closure;
843 int match = 0;
844 closure = e->multi;
845 for (i = 0; closure[i].func; i++) {
846 if (closure[i].func != probe)
847 continue;
848 if (match++ == num)
849 return closure[i].probe_private;
850 }
851 }
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700852 }
853 }
854 return ERR_PTR(-ENOENT);
855}
856EXPORT_SYMBOL_GPL(marker_get_private_data);