aboutsummaryrefslogtreecommitdiffstats
path: root/kernel
diff options
context:
space:
mode:
Diffstat (limited to 'kernel')
-rw-r--r--kernel/Kconfig.instrumentation6
-rw-r--r--kernel/Makefile1
-rw-r--r--kernel/marker.c525
-rw-r--r--kernel/module.c30
4 files changed, 562 insertions, 0 deletions
diff --git a/kernel/Kconfig.instrumentation b/kernel/Kconfig.instrumentation
index ba5c05ca2a91..f5f2c769d95e 100644
--- a/kernel/Kconfig.instrumentation
+++ b/kernel/Kconfig.instrumentation
@@ -40,4 +40,10 @@ config KPROBES
40 for kernel debugging, non-intrusive instrumentation and testing. 40 for kernel debugging, non-intrusive instrumentation and testing.
41 If in doubt, say "N". 41 If in doubt, say "N".
42 42
43config MARKERS
44 bool "Activate markers"
45 help
46 Place an empty function call at each marker site. Can be
47 dynamically changed for a probe function.
48
43endif # INSTRUMENTATION 49endif # INSTRUMENTATION
diff --git a/kernel/Makefile b/kernel/Makefile
index 32b2d8bdc9f5..05c3e6df8597 100644
--- a/kernel/Makefile
+++ b/kernel/Makefile
@@ -55,6 +55,7 @@ obj-$(CONFIG_RELAY) += relay.o
55obj-$(CONFIG_SYSCTL) += utsname_sysctl.o 55obj-$(CONFIG_SYSCTL) += utsname_sysctl.o
56obj-$(CONFIG_TASK_DELAY_ACCT) += delayacct.o 56obj-$(CONFIG_TASK_DELAY_ACCT) += delayacct.o
57obj-$(CONFIG_TASKSTATS) += taskstats.o tsacct.o 57obj-$(CONFIG_TASKSTATS) += taskstats.o tsacct.o
58obj-$(CONFIG_MARKERS) += marker.o
58 59
59ifneq ($(CONFIG_SCHED_NO_NO_OMIT_FRAME_POINTER),y) 60ifneq ($(CONFIG_SCHED_NO_NO_OMIT_FRAME_POINTER),y)
60# According to Alan Modra <alan@linuxcare.com.au>, the -fno-omit-frame-pointer is 61# According to Alan Modra <alan@linuxcare.com.au>, the -fno-omit-frame-pointer is
diff --git a/kernel/marker.c b/kernel/marker.c
new file mode 100644
index 000000000000..ccb48d9a3657
--- /dev/null
+++ b/kernel/marker.c
@@ -0,0 +1,525 @@
1/*
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>
26
27extern struct marker __start___markers[];
28extern struct marker __stop___markers[];
29
30/*
31 * module_mutex nests inside markers_mutex. Markers mutex protects the builtin
32 * and module markers, the hash table and deferred_sync.
33 */
34static DEFINE_MUTEX(markers_mutex);
35
36/*
37 * Marker deferred synchronization.
38 * Upon marker probe_unregister, we delay call to synchronize_sched() to
39 * accelerate mass unregistration (only when there is no more reference to a
40 * given module do we call synchronize_sched()). However, we need to make sure
41 * every critical region has ended before we re-arm a marker that has been
42 * unregistered and then registered back with a different probe data.
43 */
44static int deferred_sync;
45
46/*
47 * Marker hash table, containing the active markers.
48 * Protected by module_mutex.
49 */
50#define MARKER_HASH_BITS 6
51#define MARKER_TABLE_SIZE (1 << MARKER_HASH_BITS)
52
53struct marker_entry {
54 struct hlist_node hlist;
55 char *format;
56 marker_probe_func *probe;
57 void *private;
58 int refcount; /* Number of times armed. 0 if disarmed. */
59 char name[0]; /* Contains name'\0'format'\0' */
60};
61
62static struct hlist_head marker_table[MARKER_TABLE_SIZE];
63
64/**
65 * __mark_empty_function - Empty probe callback
66 * @mdata: pointer of type const struct marker
67 * @fmt: format string
68 * @...: variable argument list
69 *
70 * Empty callback provided as a probe to the markers. By providing this to a
71 * disabled marker, we make sure the execution flow is always valid even
72 * though the function pointer change and the marker enabling are two distinct
73 * operations that modifies the execution flow of preemptible code.
74 */
75void __mark_empty_function(const struct marker *mdata, void *private,
76 const char *fmt, ...)
77{
78}
79EXPORT_SYMBOL_GPL(__mark_empty_function);
80
81/*
82 * Get marker if the marker is present in the marker hash table.
83 * Must be called with markers_mutex held.
84 * Returns NULL if not present.
85 */
86static struct marker_entry *get_marker(const char *name)
87{
88 struct hlist_head *head;
89 struct hlist_node *node;
90 struct marker_entry *e;
91 u32 hash = jhash(name, strlen(name), 0);
92
93 head = &marker_table[hash & ((1 << MARKER_HASH_BITS)-1)];
94 hlist_for_each_entry(e, node, head, hlist) {
95 if (!strcmp(name, e->name))
96 return e;
97 }
98 return NULL;
99}
100
101/*
102 * Add the marker to the marker hash table. Must be called with markers_mutex
103 * held.
104 */
105static int add_marker(const char *name, const char *format,
106 marker_probe_func *probe, void *private)
107{
108 struct hlist_head *head;
109 struct hlist_node *node;
110 struct marker_entry *e;
111 size_t name_len = strlen(name) + 1;
112 size_t format_len = 0;
113 u32 hash = jhash(name, name_len-1, 0);
114
115 if (format)
116 format_len = strlen(format) + 1;
117 head = &marker_table[hash & ((1 << MARKER_HASH_BITS)-1)];
118 hlist_for_each_entry(e, node, head, hlist) {
119 if (!strcmp(name, e->name)) {
120 printk(KERN_NOTICE
121 "Marker %s busy, probe %p already installed\n",
122 name, e->probe);
123 return -EBUSY; /* Already there */
124 }
125 }
126 /*
127 * Using kmalloc here to allocate a variable length element. Could
128 * cause some memory fragmentation if overused.
129 */
130 e = kmalloc(sizeof(struct marker_entry) + name_len + format_len,
131 GFP_KERNEL);
132 if (!e)
133 return -ENOMEM;
134 memcpy(&e->name[0], name, name_len);
135 if (format) {
136 e->format = &e->name[name_len];
137 memcpy(e->format, format, format_len);
138 trace_mark(core_marker_format, "name %s format %s",
139 e->name, e->format);
140 } else
141 e->format = NULL;
142 e->probe = probe;
143 e->private = private;
144 e->refcount = 0;
145 hlist_add_head(&e->hlist, head);
146 return 0;
147}
148
149/*
150 * Remove the marker from the marker hash table. Must be called with mutex_lock
151 * held.
152 */
153static void *remove_marker(const char *name)
154{
155 struct hlist_head *head;
156 struct hlist_node *node;
157 struct marker_entry *e;
158 int found = 0;
159 size_t len = strlen(name) + 1;
160 void *private = NULL;
161 u32 hash = jhash(name, len-1, 0);
162
163 head = &marker_table[hash & ((1 << MARKER_HASH_BITS)-1)];
164 hlist_for_each_entry(e, node, head, hlist) {
165 if (!strcmp(name, e->name)) {
166 found = 1;
167 break;
168 }
169 }
170 if (found) {
171 private = e->private;
172 hlist_del(&e->hlist);
173 kfree(e);
174 }
175 return private;
176}
177
178/*
179 * Set the mark_entry format to the format found in the element.
180 */
181static int marker_set_format(struct marker_entry **entry, const char *format)
182{
183 struct marker_entry *e;
184 size_t name_len = strlen((*entry)->name) + 1;
185 size_t format_len = strlen(format) + 1;
186
187 e = kmalloc(sizeof(struct marker_entry) + name_len + format_len,
188 GFP_KERNEL);
189 if (!e)
190 return -ENOMEM;
191 memcpy(&e->name[0], (*entry)->name, name_len);
192 e->format = &e->name[name_len];
193 memcpy(e->format, format, format_len);
194 e->probe = (*entry)->probe;
195 e->private = (*entry)->private;
196 e->refcount = (*entry)->refcount;
197 hlist_add_before(&e->hlist, &(*entry)->hlist);
198 hlist_del(&(*entry)->hlist);
199 kfree(*entry);
200 *entry = e;
201 trace_mark(core_marker_format, "name %s format %s",
202 e->name, e->format);
203 return 0;
204}
205
206/*
207 * Sets the probe callback corresponding to one marker.
208 */
209static int set_marker(struct marker_entry **entry, struct marker *elem)
210{
211 int ret;
212 WARN_ON(strcmp((*entry)->name, elem->name) != 0);
213
214 if ((*entry)->format) {
215 if (strcmp((*entry)->format, elem->format) != 0) {
216 printk(KERN_NOTICE
217 "Format mismatch for probe %s "
218 "(%s), marker (%s)\n",
219 (*entry)->name,
220 (*entry)->format,
221 elem->format);
222 return -EPERM;
223 }
224 } else {
225 ret = marker_set_format(entry, elem->format);
226 if (ret)
227 return ret;
228 }
229 elem->call = (*entry)->probe;
230 elem->private = (*entry)->private;
231 elem->state = 1;
232 return 0;
233}
234
235/*
236 * Disable a marker and its probe callback.
237 * Note: only after a synchronize_sched() issued after setting elem->call to the
238 * empty function insures that the original callback is not used anymore. This
239 * insured by preemption disabling around the call site.
240 */
241static void disable_marker(struct marker *elem)
242{
243 elem->state = 0;
244 elem->call = __mark_empty_function;
245 /*
246 * Leave the private data and id there, because removal is racy and
247 * should be done only after a synchronize_sched(). These are never used
248 * until the next initialization anyway.
249 */
250}
251
252/**
253 * marker_update_probe_range - Update a probe range
254 * @begin: beginning of the range
255 * @end: end of the range
256 * @probe_module: module address of the probe being updated
257 * @refcount: number of references left to the given probe_module (out)
258 *
259 * Updates the probe callback corresponding to a range of markers.
260 * Must be called with markers_mutex held.
261 */
262void marker_update_probe_range(struct marker *begin,
263 struct marker *end, struct module *probe_module,
264 int *refcount)
265{
266 struct marker *iter;
267 struct marker_entry *mark_entry;
268
269 for (iter = begin; iter < end; iter++) {
270 mark_entry = get_marker(iter->name);
271 if (mark_entry && mark_entry->refcount) {
272 set_marker(&mark_entry, iter);
273 /*
274 * ignore error, continue
275 */
276 if (probe_module)
277 if (probe_module ==
278 __module_text_address((unsigned long)mark_entry->probe))
279 (*refcount)++;
280 } else {
281 disable_marker(iter);
282 }
283 }
284}
285
286/*
287 * Update probes, removing the faulty probes.
288 * Issues a synchronize_sched() when no reference to the module passed
289 * as parameter is found in the probes so the probe module can be
290 * safely unloaded from now on.
291 */
292static void marker_update_probes(struct module *probe_module)
293{
294 int refcount = 0;
295
296 mutex_lock(&markers_mutex);
297 /* Core kernel markers */
298 marker_update_probe_range(__start___markers,
299 __stop___markers, probe_module, &refcount);
300 /* Markers in modules. */
301 module_update_markers(probe_module, &refcount);
302 if (probe_module && refcount == 0) {
303 synchronize_sched();
304 deferred_sync = 0;
305 }
306 mutex_unlock(&markers_mutex);
307}
308
309/**
310 * marker_probe_register - Connect a probe to a marker
311 * @name: marker name
312 * @format: format string
313 * @probe: probe handler
314 * @private: probe private data
315 *
316 * private data must be a valid allocated memory address, or NULL.
317 * Returns 0 if ok, error value on error.
318 */
319int marker_probe_register(const char *name, const char *format,
320 marker_probe_func *probe, void *private)
321{
322 struct marker_entry *entry;
323 int ret = 0, need_update = 0;
324
325 mutex_lock(&markers_mutex);
326 entry = get_marker(name);
327 if (entry && entry->refcount) {
328 ret = -EBUSY;
329 goto end;
330 }
331 if (deferred_sync) {
332 synchronize_sched();
333 deferred_sync = 0;
334 }
335 ret = add_marker(name, format, probe, private);
336 if (ret)
337 goto end;
338 need_update = 1;
339end:
340 mutex_unlock(&markers_mutex);
341 if (need_update)
342 marker_update_probes(NULL);
343 return ret;
344}
345EXPORT_SYMBOL_GPL(marker_probe_register);
346
347/**
348 * marker_probe_unregister - Disconnect a probe from a marker
349 * @name: marker name
350 *
351 * Returns the private data given to marker_probe_register, or an ERR_PTR().
352 */
353void *marker_probe_unregister(const char *name)
354{
355 struct module *probe_module;
356 struct marker_entry *entry;
357 void *private;
358 int need_update = 0;
359
360 mutex_lock(&markers_mutex);
361 entry = get_marker(name);
362 if (!entry) {
363 private = ERR_PTR(-ENOENT);
364 goto end;
365 }
366 entry->refcount = 0;
367 /* In what module is the probe handler ? */
368 probe_module = __module_text_address((unsigned long)entry->probe);
369 private = remove_marker(name);
370 deferred_sync = 1;
371 need_update = 1;
372end:
373 mutex_unlock(&markers_mutex);
374 if (need_update)
375 marker_update_probes(probe_module);
376 return private;
377}
378EXPORT_SYMBOL_GPL(marker_probe_unregister);
379
380/**
381 * marker_probe_unregister_private_data - Disconnect a probe from a marker
382 * @private: probe private data
383 *
384 * Unregister a marker by providing the registered private data.
385 * Returns the private data given to marker_probe_register, or an ERR_PTR().
386 */
387void *marker_probe_unregister_private_data(void *private)
388{
389 struct module *probe_module;
390 struct hlist_head *head;
391 struct hlist_node *node;
392 struct marker_entry *entry;
393 int found = 0;
394 unsigned int i;
395 int need_update = 0;
396
397 mutex_lock(&markers_mutex);
398 for (i = 0; i < MARKER_TABLE_SIZE; i++) {
399 head = &marker_table[i];
400 hlist_for_each_entry(entry, node, head, hlist) {
401 if (entry->private == private) {
402 found = 1;
403 goto iter_end;
404 }
405 }
406 }
407iter_end:
408 if (!found) {
409 private = ERR_PTR(-ENOENT);
410 goto end;
411 }
412 entry->refcount = 0;
413 /* In what module is the probe handler ? */
414 probe_module = __module_text_address((unsigned long)entry->probe);
415 private = remove_marker(entry->name);
416 deferred_sync = 1;
417 need_update = 1;
418end:
419 mutex_unlock(&markers_mutex);
420 if (need_update)
421 marker_update_probes(probe_module);
422 return private;
423}
424EXPORT_SYMBOL_GPL(marker_probe_unregister_private_data);
425
426/**
427 * marker_arm - Arm a marker
428 * @name: marker name
429 *
430 * Activate a marker. It keeps a reference count of the number of
431 * arming/disarming done.
432 * Returns 0 if ok, error value on error.
433 */
434int marker_arm(const char *name)
435{
436 struct marker_entry *entry;
437 int ret = 0, need_update = 0;
438
439 mutex_lock(&markers_mutex);
440 entry = get_marker(name);
441 if (!entry) {
442 ret = -ENOENT;
443 goto end;
444 }
445 /*
446 * Only need to update probes when refcount passes from 0 to 1.
447 */
448 if (entry->refcount++)
449 goto end;
450 need_update = 1;
451end:
452 mutex_unlock(&markers_mutex);
453 if (need_update)
454 marker_update_probes(NULL);
455 return ret;
456}
457EXPORT_SYMBOL_GPL(marker_arm);
458
459/**
460 * marker_disarm - Disarm a marker
461 * @name: marker name
462 *
463 * Disarm a marker. It keeps a reference count of the number of arming/disarming
464 * done.
465 * Returns 0 if ok, error value on error.
466 */
467int marker_disarm(const char *name)
468{
469 struct marker_entry *entry;
470 int ret = 0, need_update = 0;
471
472 mutex_lock(&markers_mutex);
473 entry = get_marker(name);
474 if (!entry) {
475 ret = -ENOENT;
476 goto end;
477 }
478 /*
479 * Only permit decrement refcount if higher than 0.
480 * Do probe update only on 1 -> 0 transition.
481 */
482 if (entry->refcount) {
483 if (--entry->refcount)
484 goto end;
485 } else {
486 ret = -EPERM;
487 goto end;
488 }
489 need_update = 1;
490end:
491 mutex_unlock(&markers_mutex);
492 if (need_update)
493 marker_update_probes(NULL);
494 return ret;
495}
496EXPORT_SYMBOL_GPL(marker_disarm);
497
498/**
499 * marker_get_private_data - Get a marker's probe private data
500 * @name: marker name
501 *
502 * Returns the private data pointer, or an ERR_PTR.
503 * The private data pointer should _only_ be dereferenced if the caller is the
504 * owner of the data, or its content could vanish. This is mostly used to
505 * confirm that a caller is the owner of a registered probe.
506 */
507void *marker_get_private_data(const char *name)
508{
509 struct hlist_head *head;
510 struct hlist_node *node;
511 struct marker_entry *e;
512 size_t name_len = strlen(name) + 1;
513 u32 hash = jhash(name, name_len-1, 0);
514 int found = 0;
515
516 head = &marker_table[hash & ((1 << MARKER_HASH_BITS)-1)];
517 hlist_for_each_entry(e, node, head, hlist) {
518 if (!strcmp(name, e->name)) {
519 found = 1;
520 return e->private;
521 }
522 }
523 return ERR_PTR(-ENOENT);
524}
525EXPORT_SYMBOL_GPL(marker_get_private_data);
diff --git a/kernel/module.c b/kernel/module.c
index 7734595bd329..3202c9950073 100644
--- a/kernel/module.c
+++ b/kernel/module.c
@@ -1673,6 +1673,8 @@ static struct module *load_module(void __user *umod,
1673 unsigned int unusedcrcindex; 1673 unsigned int unusedcrcindex;
1674 unsigned int unusedgplindex; 1674 unsigned int unusedgplindex;
1675 unsigned int unusedgplcrcindex; 1675 unsigned int unusedgplcrcindex;
1676 unsigned int markersindex;
1677 unsigned int markersstringsindex;
1676 struct module *mod; 1678 struct module *mod;
1677 long err = 0; 1679 long err = 0;
1678 void *percpu = NULL, *ptr = NULL; /* Stops spurious gcc warning */ 1680 void *percpu = NULL, *ptr = NULL; /* Stops spurious gcc warning */
@@ -1939,6 +1941,9 @@ static struct module *load_module(void __user *umod,
1939 add_taint_module(mod, TAINT_FORCED_MODULE); 1941 add_taint_module(mod, TAINT_FORCED_MODULE);
1940 } 1942 }
1941#endif 1943#endif
1944 markersindex = find_sec(hdr, sechdrs, secstrings, "__markers");
1945 markersstringsindex = find_sec(hdr, sechdrs, secstrings,
1946 "__markers_strings");
1942 1947
1943 /* Now do relocations. */ 1948 /* Now do relocations. */
1944 for (i = 1; i < hdr->e_shnum; i++) { 1949 for (i = 1; i < hdr->e_shnum; i++) {
@@ -1961,6 +1966,11 @@ static struct module *load_module(void __user *umod,
1961 if (err < 0) 1966 if (err < 0)
1962 goto cleanup; 1967 goto cleanup;
1963 } 1968 }
1969#ifdef CONFIG_MARKERS
1970 mod->markers = (void *)sechdrs[markersindex].sh_addr;
1971 mod->num_markers =
1972 sechdrs[markersindex].sh_size / sizeof(*mod->markers);
1973#endif
1964 1974
1965 /* Find duplicate symbols */ 1975 /* Find duplicate symbols */
1966 err = verify_export_symbols(mod); 1976 err = verify_export_symbols(mod);
@@ -1979,6 +1989,11 @@ static struct module *load_module(void __user *umod,
1979 1989
1980 add_kallsyms(mod, sechdrs, symindex, strindex, secstrings); 1990 add_kallsyms(mod, sechdrs, symindex, strindex, secstrings);
1981 1991
1992#ifdef CONFIG_MARKERS
1993 if (!mod->taints)
1994 marker_update_probe_range(mod->markers,
1995 mod->markers + mod->num_markers, NULL, NULL);
1996#endif
1982 err = module_finalize(hdr, sechdrs, mod); 1997 err = module_finalize(hdr, sechdrs, mod);
1983 if (err < 0) 1998 if (err < 0)
1984 goto cleanup; 1999 goto cleanup;
@@ -2570,3 +2585,18 @@ EXPORT_SYMBOL(module_remove_driver);
2570void struct_module(struct module *mod) { return; } 2585void struct_module(struct module *mod) { return; }
2571EXPORT_SYMBOL(struct_module); 2586EXPORT_SYMBOL(struct_module);
2572#endif 2587#endif
2588
2589#ifdef CONFIG_MARKERS
2590void module_update_markers(struct module *probe_module, int *refcount)
2591{
2592 struct module *mod;
2593
2594 mutex_lock(&module_mutex);
2595 list_for_each_entry(mod, &modules, list)
2596 if (!mod->taints)
2597 marker_update_probe_range(mod->markers,
2598 mod->markers + mod->num_markers,
2599 probe_module, refcount);
2600 mutex_unlock(&module_mutex);
2601}
2602#endif