blob: 971da531790316bbe4bff400d7dec77e46c76a42 [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 /*
129 * multi points to an array, therefore accessing the array
130 * depends on reading multi. However, even in this case,
131 * we must insure that the pointer is read _before_ the array
132 * data. Same as rcu_dereference, but we need a full smp_rmb()
133 * in the fast path, so put the explicit barrier here.
134 */
135 smp_read_barrier_depends();
Mathieu Desnoyers58336112008-03-24 12:29:49 -0700136 multi = mdata->multi;
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800137 for (i = 0; multi[i].func; i++) {
Mathieu Desnoyersdc102a82008-05-12 21:21:09 +0200138 va_start(args, call_private);
139 multi[i].func(multi[i].probe_private, call_private,
140 mdata->format, &args);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800141 va_end(args);
142 }
143 }
144 preempt_enable();
145}
146EXPORT_SYMBOL_GPL(marker_probe_cb);
147
148/*
149 * marker_probe_cb Callback that does not prepare the variable argument list.
150 * @mdata: pointer of type struct marker
151 * @call_private: caller site private data
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800152 * @...: Variable argument list.
153 *
154 * Should be connected to markers "MARK_NOARGS".
155 */
Mathieu Desnoyersdc102a82008-05-12 21:21:09 +0200156void marker_probe_cb_noarg(const struct marker *mdata, void *call_private, ...)
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800157{
158 va_list args; /* not initialized */
159 char ptype;
160
161 preempt_disable();
Mathieu Desnoyers58336112008-03-24 12:29:49 -0700162 ptype = mdata->ptype;
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800163 if (likely(!ptype)) {
164 marker_probe_func *func;
165 /* Must read the ptype before ptr. They are not data dependant,
166 * so we put an explicit smp_rmb() here. */
167 smp_rmb();
Mathieu Desnoyers58336112008-03-24 12:29:49 -0700168 func = mdata->single.func;
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800169 /* Must read the ptr before private data. They are not data
170 * dependant, so we put an explicit smp_rmb() here. */
171 smp_rmb();
Mathieu Desnoyersdc102a82008-05-12 21:21:09 +0200172 func(mdata->single.probe_private, call_private, mdata->format,
173 &args);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800174 } else {
175 struct marker_probe_closure *multi;
176 int i;
177 /*
178 * multi points to an array, therefore accessing the array
179 * depends on reading multi. However, even in this case,
180 * we must insure that the pointer is read _before_ the array
181 * data. Same as rcu_dereference, but we need a full smp_rmb()
182 * in the fast path, so put the explicit barrier here.
183 */
184 smp_read_barrier_depends();
Mathieu Desnoyers58336112008-03-24 12:29:49 -0700185 multi = mdata->multi;
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800186 for (i = 0; multi[i].func; i++)
Mathieu Desnoyersdc102a82008-05-12 21:21:09 +0200187 multi[i].func(multi[i].probe_private, call_private,
188 mdata->format, &args);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800189 }
190 preempt_enable();
191}
192EXPORT_SYMBOL_GPL(marker_probe_cb_noarg);
193
194static void free_old_closure(struct rcu_head *head)
195{
196 struct marker_entry *entry = container_of(head,
197 struct marker_entry, rcu);
198 kfree(entry->oldptr);
199 /* Make sure we free the data before setting the pending flag to 0 */
200 smp_wmb();
201 entry->rcu_pending = 0;
202}
203
204static void debug_print_probes(struct marker_entry *entry)
205{
206 int i;
207
208 if (!marker_debug)
209 return;
210
211 if (!entry->ptype) {
212 printk(KERN_DEBUG "Single probe : %p %p\n",
213 entry->single.func,
214 entry->single.probe_private);
215 } else {
216 for (i = 0; entry->multi[i].func; i++)
217 printk(KERN_DEBUG "Multi probe %d : %p %p\n", i,
218 entry->multi[i].func,
219 entry->multi[i].probe_private);
220 }
221}
222
223static struct marker_probe_closure *
224marker_entry_add_probe(struct marker_entry *entry,
225 marker_probe_func *probe, void *probe_private)
226{
227 int nr_probes = 0;
228 struct marker_probe_closure *old, *new;
229
230 WARN_ON(!probe);
231
232 debug_print_probes(entry);
233 old = entry->multi;
234 if (!entry->ptype) {
235 if (entry->single.func == probe &&
236 entry->single.probe_private == probe_private)
237 return ERR_PTR(-EBUSY);
238 if (entry->single.func == __mark_empty_function) {
239 /* 0 -> 1 probes */
240 entry->single.func = probe;
241 entry->single.probe_private = probe_private;
242 entry->refcount = 1;
243 entry->ptype = 0;
244 debug_print_probes(entry);
245 return NULL;
246 } else {
247 /* 1 -> 2 probes */
248 nr_probes = 1;
249 old = NULL;
250 }
251 } else {
252 /* (N -> N+1), (N != 0, 1) probes */
253 for (nr_probes = 0; old[nr_probes].func; nr_probes++)
254 if (old[nr_probes].func == probe
255 && old[nr_probes].probe_private
256 == probe_private)
257 return ERR_PTR(-EBUSY);
258 }
259 /* + 2 : one for new probe, one for NULL func */
260 new = kzalloc((nr_probes + 2) * sizeof(struct marker_probe_closure),
261 GFP_KERNEL);
262 if (new == NULL)
263 return ERR_PTR(-ENOMEM);
264 if (!old)
265 new[0] = entry->single;
266 else
267 memcpy(new, old,
268 nr_probes * sizeof(struct marker_probe_closure));
269 new[nr_probes].func = probe;
270 new[nr_probes].probe_private = probe_private;
271 entry->refcount = nr_probes + 1;
272 entry->multi = new;
273 entry->ptype = 1;
274 debug_print_probes(entry);
275 return old;
276}
277
278static struct marker_probe_closure *
279marker_entry_remove_probe(struct marker_entry *entry,
280 marker_probe_func *probe, void *probe_private)
281{
282 int nr_probes = 0, nr_del = 0, i;
283 struct marker_probe_closure *old, *new;
284
285 old = entry->multi;
286
287 debug_print_probes(entry);
288 if (!entry->ptype) {
289 /* 0 -> N is an error */
290 WARN_ON(entry->single.func == __mark_empty_function);
291 /* 1 -> 0 probes */
292 WARN_ON(probe && entry->single.func != probe);
293 WARN_ON(entry->single.probe_private != probe_private);
294 entry->single.func = __mark_empty_function;
295 entry->refcount = 0;
296 entry->ptype = 0;
297 debug_print_probes(entry);
298 return NULL;
299 } else {
300 /* (N -> M), (N > 1, M >= 0) probes */
301 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
302 if ((!probe || old[nr_probes].func == probe)
303 && old[nr_probes].probe_private
304 == probe_private)
305 nr_del++;
306 }
307 }
308
309 if (nr_probes - nr_del == 0) {
310 /* N -> 0, (N > 1) */
311 entry->single.func = __mark_empty_function;
312 entry->refcount = 0;
313 entry->ptype = 0;
314 } else if (nr_probes - nr_del == 1) {
315 /* N -> 1, (N > 1) */
316 for (i = 0; old[i].func; i++)
317 if ((probe && old[i].func != probe) ||
318 old[i].probe_private != probe_private)
319 entry->single = old[i];
320 entry->refcount = 1;
321 entry->ptype = 0;
322 } else {
323 int j = 0;
324 /* N -> M, (N > 1, M > 1) */
325 /* + 1 for NULL */
326 new = kzalloc((nr_probes - nr_del + 1)
327 * sizeof(struct marker_probe_closure), GFP_KERNEL);
328 if (new == NULL)
329 return ERR_PTR(-ENOMEM);
330 for (i = 0; old[i].func; i++)
331 if ((probe && old[i].func != probe) ||
332 old[i].probe_private != probe_private)
333 new[j++] = old[i];
334 entry->refcount = nr_probes - nr_del;
335 entry->ptype = 1;
336 entry->multi = new;
337 }
338 debug_print_probes(entry);
339 return old;
340}
341
342/*
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700343 * Get marker if the marker is present in the marker hash table.
344 * Must be called with markers_mutex held.
345 * Returns NULL if not present.
346 */
347static struct marker_entry *get_marker(const char *name)
348{
349 struct hlist_head *head;
350 struct hlist_node *node;
351 struct marker_entry *e;
352 u32 hash = jhash(name, strlen(name), 0);
353
354 head = &marker_table[hash & ((1 << MARKER_HASH_BITS)-1)];
355 hlist_for_each_entry(e, node, head, hlist) {
356 if (!strcmp(name, e->name))
357 return e;
358 }
359 return NULL;
360}
361
362/*
363 * Add the marker to the marker hash table. Must be called with markers_mutex
364 * held.
365 */
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800366static struct marker_entry *add_marker(const char *name, const char *format)
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700367{
368 struct hlist_head *head;
369 struct hlist_node *node;
370 struct marker_entry *e;
371 size_t name_len = strlen(name) + 1;
372 size_t format_len = 0;
373 u32 hash = jhash(name, name_len-1, 0);
374
375 if (format)
376 format_len = strlen(format) + 1;
377 head = &marker_table[hash & ((1 << MARKER_HASH_BITS)-1)];
378 hlist_for_each_entry(e, node, head, hlist) {
379 if (!strcmp(name, e->name)) {
380 printk(KERN_NOTICE
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800381 "Marker %s busy\n", name);
382 return ERR_PTR(-EBUSY); /* Already there */
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700383 }
384 }
385 /*
386 * Using kmalloc here to allocate a variable length element. Could
387 * cause some memory fragmentation if overused.
388 */
389 e = kmalloc(sizeof(struct marker_entry) + name_len + format_len,
390 GFP_KERNEL);
391 if (!e)
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800392 return ERR_PTR(-ENOMEM);
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700393 memcpy(&e->name[0], name, name_len);
394 if (format) {
395 e->format = &e->name[name_len];
396 memcpy(e->format, format, format_len);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800397 if (strcmp(e->format, MARK_NOARGS) == 0)
398 e->call = marker_probe_cb_noarg;
399 else
400 e->call = marker_probe_cb;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700401 trace_mark(core_marker_format, "name %s format %s",
402 e->name, e->format);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800403 } else {
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700404 e->format = NULL;
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800405 e->call = marker_probe_cb;
406 }
407 e->single.func = __mark_empty_function;
408 e->single.probe_private = NULL;
409 e->multi = NULL;
410 e->ptype = 0;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700411 e->refcount = 0;
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800412 e->rcu_pending = 0;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700413 hlist_add_head(&e->hlist, head);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800414 return e;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700415}
416
417/*
418 * Remove the marker from the marker hash table. Must be called with mutex_lock
419 * held.
420 */
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800421static int remove_marker(const char *name)
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700422{
423 struct hlist_head *head;
424 struct hlist_node *node;
425 struct marker_entry *e;
426 int found = 0;
427 size_t len = strlen(name) + 1;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700428 u32 hash = jhash(name, len-1, 0);
429
430 head = &marker_table[hash & ((1 << MARKER_HASH_BITS)-1)];
431 hlist_for_each_entry(e, node, head, hlist) {
432 if (!strcmp(name, e->name)) {
433 found = 1;
434 break;
435 }
436 }
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800437 if (!found)
438 return -ENOENT;
439 if (e->single.func != __mark_empty_function)
440 return -EBUSY;
441 hlist_del(&e->hlist);
442 /* Make sure the call_rcu has been executed */
443 if (e->rcu_pending)
Mathieu Desnoyers28325df2008-07-25 01:48:38 -0700444 rcu_barrier_sched();
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800445 kfree(e);
446 return 0;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700447}
448
449/*
450 * Set the mark_entry format to the format found in the element.
451 */
452static int marker_set_format(struct marker_entry **entry, const char *format)
453{
454 struct marker_entry *e;
455 size_t name_len = strlen((*entry)->name) + 1;
456 size_t format_len = strlen(format) + 1;
457
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800458
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700459 e = kmalloc(sizeof(struct marker_entry) + name_len + format_len,
460 GFP_KERNEL);
461 if (!e)
462 return -ENOMEM;
463 memcpy(&e->name[0], (*entry)->name, name_len);
464 e->format = &e->name[name_len];
465 memcpy(e->format, format, format_len);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800466 if (strcmp(e->format, MARK_NOARGS) == 0)
467 e->call = marker_probe_cb_noarg;
468 else
469 e->call = marker_probe_cb;
470 e->single = (*entry)->single;
471 e->multi = (*entry)->multi;
472 e->ptype = (*entry)->ptype;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700473 e->refcount = (*entry)->refcount;
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800474 e->rcu_pending = 0;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700475 hlist_add_before(&e->hlist, &(*entry)->hlist);
476 hlist_del(&(*entry)->hlist);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800477 /* Make sure the call_rcu has been executed */
478 if ((*entry)->rcu_pending)
Mathieu Desnoyers28325df2008-07-25 01:48:38 -0700479 rcu_barrier_sched();
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700480 kfree(*entry);
481 *entry = e;
482 trace_mark(core_marker_format, "name %s format %s",
483 e->name, e->format);
484 return 0;
485}
486
487/*
488 * Sets the probe callback corresponding to one marker.
489 */
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800490static int set_marker(struct marker_entry **entry, struct marker *elem,
491 int active)
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700492{
493 int ret;
494 WARN_ON(strcmp((*entry)->name, elem->name) != 0);
495
496 if ((*entry)->format) {
497 if (strcmp((*entry)->format, elem->format) != 0) {
498 printk(KERN_NOTICE
499 "Format mismatch for probe %s "
500 "(%s), marker (%s)\n",
501 (*entry)->name,
502 (*entry)->format,
503 elem->format);
504 return -EPERM;
505 }
506 } else {
507 ret = marker_set_format(entry, elem->format);
508 if (ret)
509 return ret;
510 }
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800511
512 /*
513 * probe_cb setup (statically known) is done here. It is
514 * asynchronous with the rest of execution, therefore we only
515 * pass from a "safe" callback (with argument) to an "unsafe"
516 * callback (does not set arguments).
517 */
518 elem->call = (*entry)->call;
519 /*
520 * Sanity check :
521 * We only update the single probe private data when the ptr is
522 * set to a _non_ single probe! (0 -> 1 and N -> 1, N != 1)
523 */
524 WARN_ON(elem->single.func != __mark_empty_function
525 && elem->single.probe_private
526 != (*entry)->single.probe_private &&
527 !elem->ptype);
528 elem->single.probe_private = (*entry)->single.probe_private;
529 /*
530 * Make sure the private data is valid when we update the
531 * single probe ptr.
532 */
533 smp_wmb();
534 elem->single.func = (*entry)->single.func;
535 /*
536 * We also make sure that the new probe callbacks array is consistent
537 * before setting a pointer to it.
538 */
539 rcu_assign_pointer(elem->multi, (*entry)->multi);
540 /*
541 * Update the function or multi probe array pointer before setting the
542 * ptype.
543 */
544 smp_wmb();
545 elem->ptype = (*entry)->ptype;
546 elem->state = active;
547
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700548 return 0;
549}
550
551/*
552 * Disable a marker and its probe callback.
Mathieu Desnoyersfd3c36f2008-03-24 12:29:47 -0700553 * Note: only waiting an RCU period after setting elem->call to the empty
554 * function insures that the original callback is not used anymore. This insured
555 * by preempt_disable around the call site.
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700556 */
557static void disable_marker(struct marker *elem)
558{
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800559 /* leave "call" as is. It is known statically. */
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700560 elem->state = 0;
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800561 elem->single.func = __mark_empty_function;
562 /* Update the function before setting the ptype */
563 smp_wmb();
564 elem->ptype = 0; /* single probe */
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700565 /*
566 * Leave the private data and id there, because removal is racy and
Mathieu Desnoyersfd3c36f2008-03-24 12:29:47 -0700567 * should be done only after an RCU period. These are never used until
568 * the next initialization anyway.
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700569 */
570}
571
572/**
573 * marker_update_probe_range - Update a probe range
574 * @begin: beginning of the range
575 * @end: end of the range
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700576 *
577 * Updates the probe callback corresponding to a range of markers.
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700578 */
579void marker_update_probe_range(struct marker *begin,
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800580 struct marker *end)
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700581{
582 struct marker *iter;
583 struct marker_entry *mark_entry;
584
Mathieu Desnoyers314de8a2007-11-14 16:59:48 -0800585 mutex_lock(&markers_mutex);
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700586 for (iter = begin; iter < end; iter++) {
587 mark_entry = get_marker(iter->name);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800588 if (mark_entry) {
589 set_marker(&mark_entry, iter,
590 !!mark_entry->refcount);
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700591 /*
592 * ignore error, continue
593 */
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700594 } else {
595 disable_marker(iter);
596 }
597 }
Mathieu Desnoyers314de8a2007-11-14 16:59:48 -0800598 mutex_unlock(&markers_mutex);
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700599}
600
601/*
602 * Update probes, removing the faulty probes.
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800603 *
604 * Internal callback only changed before the first probe is connected to it.
605 * Single probe private data can only be changed on 0 -> 1 and 2 -> 1
606 * transitions. All other transitions will leave the old private data valid.
607 * This makes the non-atomicity of the callback/private data updates valid.
608 *
609 * "special case" updates :
610 * 0 -> 1 callback
611 * 1 -> 0 callback
612 * 1 -> 2 callbacks
613 * 2 -> 1 callbacks
614 * Other updates all behave the same, just like the 2 -> 3 or 3 -> 2 updates.
615 * Site effect : marker_set_format may delete the marker entry (creating a
616 * replacement).
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700617 */
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800618static void marker_update_probes(void)
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700619{
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700620 /* Core kernel markers */
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800621 marker_update_probe_range(__start___markers, __stop___markers);
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700622 /* Markers in modules. */
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800623 module_update_markers();
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700624}
625
626/**
627 * marker_probe_register - Connect a probe to a marker
628 * @name: marker name
629 * @format: format string
630 * @probe: probe handler
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800631 * @probe_private: probe private data
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700632 *
633 * private data must be a valid allocated memory address, or NULL.
634 * Returns 0 if ok, error value on error.
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800635 * The probe address must at least be aligned on the architecture pointer size.
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700636 */
637int marker_probe_register(const char *name, const char *format,
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800638 marker_probe_func *probe, void *probe_private)
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700639{
640 struct marker_entry *entry;
Mathieu Desnoyers314de8a2007-11-14 16:59:48 -0800641 int ret = 0;
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800642 struct marker_probe_closure *old;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700643
644 mutex_lock(&markers_mutex);
645 entry = get_marker(name);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800646 if (!entry) {
647 entry = add_marker(name, format);
648 if (IS_ERR(entry)) {
649 ret = PTR_ERR(entry);
650 goto end;
651 }
652 }
653 /*
654 * If we detect that a call_rcu is pending for this marker,
655 * make sure it's executed now.
656 */
657 if (entry->rcu_pending)
Mathieu Desnoyers28325df2008-07-25 01:48:38 -0700658 rcu_barrier_sched();
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800659 old = marker_entry_add_probe(entry, probe, probe_private);
660 if (IS_ERR(old)) {
661 ret = PTR_ERR(old);
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700662 goto end;
663 }
Mathieu Desnoyers314de8a2007-11-14 16:59:48 -0800664 mutex_unlock(&markers_mutex);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800665 marker_update_probes(); /* may update entry */
666 mutex_lock(&markers_mutex);
667 entry = get_marker(name);
668 WARN_ON(!entry);
669 entry->oldptr = old;
670 entry->rcu_pending = 1;
671 /* write rcu_pending before calling the RCU callback */
672 smp_wmb();
Mathieu Desnoyers28325df2008-07-25 01:48:38 -0700673 call_rcu_sched(&entry->rcu, free_old_closure);
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700674end:
675 mutex_unlock(&markers_mutex);
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700676 return ret;
677}
678EXPORT_SYMBOL_GPL(marker_probe_register);
679
680/**
681 * marker_probe_unregister - Disconnect a probe from a marker
682 * @name: marker name
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800683 * @probe: probe function pointer
684 * @probe_private: probe private data
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700685 *
686 * Returns the private data given to marker_probe_register, or an ERR_PTR().
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800687 * We do not need to call a synchronize_sched to make sure the probes have
688 * finished running before doing a module unload, because the module unload
689 * itself uses stop_machine(), which insures that every preempt disabled section
690 * have finished.
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700691 */
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800692int marker_probe_unregister(const char *name,
693 marker_probe_func *probe, void *probe_private)
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700694{
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700695 struct marker_entry *entry;
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800696 struct marker_probe_closure *old;
Jesper Juhl544adb42008-03-04 14:29:00 -0800697 int ret = -ENOENT;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700698
699 mutex_lock(&markers_mutex);
700 entry = get_marker(name);
Jesper Juhl544adb42008-03-04 14:29:00 -0800701 if (!entry)
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700702 goto end;
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800703 if (entry->rcu_pending)
Mathieu Desnoyers28325df2008-07-25 01:48:38 -0700704 rcu_barrier_sched();
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800705 old = marker_entry_remove_probe(entry, probe, probe_private);
Mathieu Desnoyers314de8a2007-11-14 16:59:48 -0800706 mutex_unlock(&markers_mutex);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800707 marker_update_probes(); /* may update entry */
708 mutex_lock(&markers_mutex);
709 entry = get_marker(name);
Jesper Juhl544adb42008-03-04 14:29:00 -0800710 if (!entry)
711 goto end;
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800712 entry->oldptr = old;
713 entry->rcu_pending = 1;
714 /* write rcu_pending before calling the RCU callback */
715 smp_wmb();
Mathieu Desnoyers28325df2008-07-25 01:48:38 -0700716 call_rcu_sched(&entry->rcu, free_old_closure);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800717 remove_marker(name); /* Ignore busy error message */
Jesper Juhl544adb42008-03-04 14:29:00 -0800718 ret = 0;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700719end:
720 mutex_unlock(&markers_mutex);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800721 return ret;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700722}
723EXPORT_SYMBOL_GPL(marker_probe_unregister);
724
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800725static struct marker_entry *
726get_marker_from_private_data(marker_probe_func *probe, void *probe_private)
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700727{
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800728 struct marker_entry *entry;
729 unsigned int i;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700730 struct hlist_head *head;
731 struct hlist_node *node;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700732
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700733 for (i = 0; i < MARKER_TABLE_SIZE; i++) {
734 head = &marker_table[i];
735 hlist_for_each_entry(entry, node, head, hlist) {
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800736 if (!entry->ptype) {
737 if (entry->single.func == probe
738 && entry->single.probe_private
739 == probe_private)
740 return entry;
741 } else {
742 struct marker_probe_closure *closure;
743 closure = entry->multi;
744 for (i = 0; closure[i].func; i++) {
745 if (closure[i].func == probe &&
746 closure[i].probe_private
747 == probe_private)
748 return entry;
749 }
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700750 }
751 }
752 }
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800753 return NULL;
754}
755
756/**
757 * marker_probe_unregister_private_data - Disconnect a probe from a marker
758 * @probe: probe function
759 * @probe_private: probe private data
760 *
761 * Unregister a probe by providing the registered private data.
762 * Only removes the first marker found in hash table.
763 * Return 0 on success or error value.
764 * We do not need to call a synchronize_sched to make sure the probes have
765 * finished running before doing a module unload, because the module unload
766 * itself uses stop_machine(), which insures that every preempt disabled section
767 * have finished.
768 */
769int marker_probe_unregister_private_data(marker_probe_func *probe,
770 void *probe_private)
771{
772 struct marker_entry *entry;
773 int ret = 0;
774 struct marker_probe_closure *old;
775
776 mutex_lock(&markers_mutex);
777 entry = get_marker_from_private_data(probe, probe_private);
778 if (!entry) {
779 ret = -ENOENT;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700780 goto end;
781 }
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800782 if (entry->rcu_pending)
Mathieu Desnoyers28325df2008-07-25 01:48:38 -0700783 rcu_barrier_sched();
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800784 old = marker_entry_remove_probe(entry, NULL, probe_private);
Mathieu Desnoyers314de8a2007-11-14 16:59:48 -0800785 mutex_unlock(&markers_mutex);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800786 marker_update_probes(); /* may update entry */
787 mutex_lock(&markers_mutex);
788 entry = get_marker_from_private_data(probe, probe_private);
789 WARN_ON(!entry);
790 entry->oldptr = old;
791 entry->rcu_pending = 1;
792 /* write rcu_pending before calling the RCU callback */
793 smp_wmb();
Mathieu Desnoyers28325df2008-07-25 01:48:38 -0700794 call_rcu_sched(&entry->rcu, free_old_closure);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800795 remove_marker(entry->name); /* Ignore busy error message */
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700796end:
797 mutex_unlock(&markers_mutex);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800798 return ret;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700799}
800EXPORT_SYMBOL_GPL(marker_probe_unregister_private_data);
801
802/**
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700803 * marker_get_private_data - Get a marker's probe private data
804 * @name: marker name
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800805 * @probe: probe to match
806 * @num: get the nth matching probe's private data
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700807 *
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800808 * Returns the nth private data pointer (starting from 0) matching, or an
809 * ERR_PTR.
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700810 * Returns the private data pointer, or an ERR_PTR.
811 * The private data pointer should _only_ be dereferenced if the caller is the
812 * owner of the data, or its content could vanish. This is mostly used to
813 * confirm that a caller is the owner of a registered probe.
814 */
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800815void *marker_get_private_data(const char *name, marker_probe_func *probe,
816 int num)
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700817{
818 struct hlist_head *head;
819 struct hlist_node *node;
820 struct marker_entry *e;
821 size_t name_len = strlen(name) + 1;
822 u32 hash = jhash(name, name_len-1, 0);
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800823 int i;
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700824
825 head = &marker_table[hash & ((1 << MARKER_HASH_BITS)-1)];
826 hlist_for_each_entry(e, node, head, hlist) {
827 if (!strcmp(name, e->name)) {
Mathieu Desnoyersfb40bd72008-02-13 15:03:37 -0800828 if (!e->ptype) {
829 if (num == 0 && e->single.func == probe)
830 return e->single.probe_private;
831 else
832 break;
833 } else {
834 struct marker_probe_closure *closure;
835 int match = 0;
836 closure = e->multi;
837 for (i = 0; closure[i].func; i++) {
838 if (closure[i].func != probe)
839 continue;
840 if (match++ == num)
841 return closure[i].probe_private;
842 }
843 }
Mathieu Desnoyers8256e472007-10-18 23:41:06 -0700844 }
845 }
846 return ERR_PTR(-ENOENT);
847}
848EXPORT_SYMBOL_GPL(marker_get_private_data);