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 47 /* 48 * Note about RCU : 49 * It is used to to delay the free of multiple probes array until a quiescent 50 * state is reached. 51 * Tracepoint entries modifications are protected by the tracepoints_mutex. 52 */ 53 struct tracepoint_entry { 54 struct hlist_node hlist; 55 void **funcs; 56 int refcount; /* Number of times armed. 0 if disarmed. */ 57 struct rcu_head rcu; 58 void *oldptr; 59 unsigned char rcu_pending:1; 60 char name[0]; 61 }; 62 63 static struct hlist_head tracepoint_table[TRACEPOINT_TABLE_SIZE]; 64 65 static void free_old_closure(struct rcu_head *head) 66 { 67 struct tracepoint_entry *entry = container_of(head, 68 struct tracepoint_entry, rcu); 69 kfree(entry->oldptr); 70 /* Make sure we free the data before setting the pending flag to 0 */ 71 smp_wmb(); 72 entry->rcu_pending = 0; 73 } 74 75 static void tracepoint_entry_free_old(struct tracepoint_entry *entry, void *old) 76 { 77 if (!old) 78 return; 79 entry->oldptr = old; 80 entry->rcu_pending = 1; 81 /* write rcu_pending before calling the RCU callback */ 82 smp_wmb(); 83 call_rcu_sched(&entry->rcu, free_old_closure); 84 } 85 86 static void debug_print_probes(struct tracepoint_entry *entry) 87 { 88 int i; 89 90 if (!tracepoint_debug) 91 return; 92 93 for (i = 0; entry->funcs[i]; i++) 94 printk(KERN_DEBUG "Probe %d : %p\n", i, entry->funcs[i]); 95 } 96 97 static void * 98 tracepoint_entry_add_probe(struct tracepoint_entry *entry, void *probe) 99 { 100 int nr_probes = 0; 101 void **old, **new; 102 103 WARN_ON(!probe); 104 105 debug_print_probes(entry); 106 old = entry->funcs; 107 if (old) { 108 /* (N -> N+1), (N != 0, 1) probes */ 109 for (nr_probes = 0; old[nr_probes]; nr_probes++) 110 if (old[nr_probes] == probe) 111 return ERR_PTR(-EEXIST); 112 } 113 /* + 2 : one for new probe, one for NULL func */ 114 new = kzalloc((nr_probes + 2) * sizeof(void *), GFP_KERNEL); 115 if (new == NULL) 116 return ERR_PTR(-ENOMEM); 117 if (old) 118 memcpy(new, old, nr_probes * sizeof(void *)); 119 new[nr_probes] = probe; 120 entry->refcount = nr_probes + 1; 121 entry->funcs = new; 122 debug_print_probes(entry); 123 return old; 124 } 125 126 static void * 127 tracepoint_entry_remove_probe(struct tracepoint_entry *entry, void *probe) 128 { 129 int nr_probes = 0, nr_del = 0, i; 130 void **old, **new; 131 132 old = entry->funcs; 133 134 if (!old) 135 return NULL; 136 137 debug_print_probes(entry); 138 /* (N -> M), (N > 1, M >= 0) probes */ 139 for (nr_probes = 0; old[nr_probes]; nr_probes++) { 140 if ((!probe || old[nr_probes] == probe)) 141 nr_del++; 142 } 143 144 if (nr_probes - nr_del == 0) { 145 /* N -> 0, (N > 1) */ 146 entry->funcs = NULL; 147 entry->refcount = 0; 148 debug_print_probes(entry); 149 return old; 150 } else { 151 int j = 0; 152 /* N -> M, (N > 1, M > 0) */ 153 /* + 1 for NULL */ 154 new = kzalloc((nr_probes - nr_del + 1) 155 * sizeof(void *), GFP_KERNEL); 156 if (new == NULL) 157 return ERR_PTR(-ENOMEM); 158 for (i = 0; old[i]; i++) 159 if ((probe && old[i] != probe)) 160 new[j++] = old[i]; 161 entry->refcount = nr_probes - nr_del; 162 entry->funcs = new; 163 } 164 debug_print_probes(entry); 165 return old; 166 } 167 168 /* 169 * Get tracepoint if the tracepoint is present in the tracepoint hash table. 170 * Must be called with tracepoints_mutex held. 171 * Returns NULL if not present. 172 */ 173 static struct tracepoint_entry *get_tracepoint(const char *name) 174 { 175 struct hlist_head *head; 176 struct hlist_node *node; 177 struct tracepoint_entry *e; 178 u32 hash = jhash(name, strlen(name), 0); 179 180 head = &tracepoint_table[hash & (TRACEPOINT_TABLE_SIZE - 1)]; 181 hlist_for_each_entry(e, node, head, hlist) { 182 if (!strcmp(name, e->name)) 183 return e; 184 } 185 return NULL; 186 } 187 188 /* 189 * Add the tracepoint to the tracepoint hash table. Must be called with 190 * tracepoints_mutex held. 191 */ 192 static struct tracepoint_entry *add_tracepoint(const char *name) 193 { 194 struct hlist_head *head; 195 struct hlist_node *node; 196 struct tracepoint_entry *e; 197 size_t name_len = strlen(name) + 1; 198 u32 hash = jhash(name, name_len-1, 0); 199 200 head = &tracepoint_table[hash & (TRACEPOINT_TABLE_SIZE - 1)]; 201 hlist_for_each_entry(e, node, head, hlist) { 202 if (!strcmp(name, e->name)) { 203 printk(KERN_NOTICE 204 "tracepoint %s busy\n", name); 205 return ERR_PTR(-EEXIST); /* Already there */ 206 } 207 } 208 /* 209 * Using kmalloc here to allocate a variable length element. Could 210 * cause some memory fragmentation if overused. 211 */ 212 e = kmalloc(sizeof(struct tracepoint_entry) + name_len, GFP_KERNEL); 213 if (!e) 214 return ERR_PTR(-ENOMEM); 215 memcpy(&e->name[0], name, name_len); 216 e->funcs = NULL; 217 e->refcount = 0; 218 e->rcu_pending = 0; 219 hlist_add_head(&e->hlist, head); 220 return e; 221 } 222 223 /* 224 * Remove the tracepoint from the tracepoint hash table. Must be called with 225 * mutex_lock held. 226 */ 227 static int remove_tracepoint(const char *name) 228 { 229 struct hlist_head *head; 230 struct hlist_node *node; 231 struct tracepoint_entry *e; 232 int found = 0; 233 size_t len = strlen(name) + 1; 234 u32 hash = jhash(name, len-1, 0); 235 236 head = &tracepoint_table[hash & (TRACEPOINT_TABLE_SIZE - 1)]; 237 hlist_for_each_entry(e, node, head, hlist) { 238 if (!strcmp(name, e->name)) { 239 found = 1; 240 break; 241 } 242 } 243 if (!found) 244 return -ENOENT; 245 if (e->refcount) 246 return -EBUSY; 247 hlist_del(&e->hlist); 248 /* Make sure the call_rcu_sched has been executed */ 249 if (e->rcu_pending) 250 rcu_barrier_sched(); 251 kfree(e); 252 return 0; 253 } 254 255 /* 256 * Sets the probe callback corresponding to one tracepoint. 257 */ 258 static void set_tracepoint(struct tracepoint_entry **entry, 259 struct tracepoint *elem, int active) 260 { 261 WARN_ON(strcmp((*entry)->name, elem->name) != 0); 262 263 /* 264 * rcu_assign_pointer has a smp_wmb() which makes sure that the new 265 * probe callbacks array is consistent before setting a pointer to it. 266 * This array is referenced by __DO_TRACE from 267 * include/linux/tracepoints.h. A matching smp_read_barrier_depends() 268 * is used. 269 */ 270 rcu_assign_pointer(elem->funcs, (*entry)->funcs); 271 elem->state = active; 272 } 273 274 /* 275 * Disable a tracepoint and its probe callback. 276 * Note: only waiting an RCU period after setting elem->call to the empty 277 * function insures that the original callback is not used anymore. This insured 278 * by preempt_disable around the call site. 279 */ 280 static void disable_tracepoint(struct tracepoint *elem) 281 { 282 elem->state = 0; 283 } 284 285 /** 286 * tracepoint_update_probe_range - Update a probe range 287 * @begin: beginning of the range 288 * @end: end of the range 289 * 290 * Updates the probe callback corresponding to a range of tracepoints. 291 */ 292 void tracepoint_update_probe_range(struct tracepoint *begin, 293 struct tracepoint *end) 294 { 295 struct tracepoint *iter; 296 struct tracepoint_entry *mark_entry; 297 298 mutex_lock(&tracepoints_mutex); 299 for (iter = begin; iter < end; iter++) { 300 mark_entry = get_tracepoint(iter->name); 301 if (mark_entry) { 302 set_tracepoint(&mark_entry, iter, 303 !!mark_entry->refcount); 304 } else { 305 disable_tracepoint(iter); 306 } 307 } 308 mutex_unlock(&tracepoints_mutex); 309 } 310 311 /* 312 * Update probes, removing the faulty probes. 313 */ 314 static void tracepoint_update_probes(void) 315 { 316 /* Core kernel tracepoints */ 317 tracepoint_update_probe_range(__start___tracepoints, 318 __stop___tracepoints); 319 /* tracepoints in modules. */ 320 module_update_tracepoints(); 321 } 322 323 /** 324 * tracepoint_probe_register - Connect a probe to a tracepoint 325 * @name: tracepoint name 326 * @probe: probe handler 327 * 328 * Returns 0 if ok, error value on error. 329 * The probe address must at least be aligned on the architecture pointer size. 330 */ 331 int tracepoint_probe_register(const char *name, void *probe) 332 { 333 struct tracepoint_entry *entry; 334 int ret = 0; 335 void *old; 336 337 mutex_lock(&tracepoints_mutex); 338 entry = get_tracepoint(name); 339 if (!entry) { 340 entry = add_tracepoint(name); 341 if (IS_ERR(entry)) { 342 ret = PTR_ERR(entry); 343 goto end; 344 } 345 } 346 /* 347 * If we detect that a call_rcu_sched is pending for this tracepoint, 348 * make sure it's executed now. 349 */ 350 if (entry->rcu_pending) 351 rcu_barrier_sched(); 352 old = tracepoint_entry_add_probe(entry, probe); 353 if (IS_ERR(old)) { 354 ret = PTR_ERR(old); 355 goto end; 356 } 357 mutex_unlock(&tracepoints_mutex); 358 tracepoint_update_probes(); /* may update entry */ 359 mutex_lock(&tracepoints_mutex); 360 entry = get_tracepoint(name); 361 WARN_ON(!entry); 362 if (entry->rcu_pending) 363 rcu_barrier_sched(); 364 tracepoint_entry_free_old(entry, old); 365 end: 366 mutex_unlock(&tracepoints_mutex); 367 return ret; 368 } 369 EXPORT_SYMBOL_GPL(tracepoint_probe_register); 370 371 /** 372 * tracepoint_probe_unregister - Disconnect a probe from a tracepoint 373 * @name: tracepoint name 374 * @probe: probe function pointer 375 * 376 * We do not need to call a synchronize_sched to make sure the probes have 377 * finished running before doing a module unload, because the module unload 378 * itself uses stop_machine(), which insures that every preempt disabled section 379 * have finished. 380 */ 381 int tracepoint_probe_unregister(const char *name, void *probe) 382 { 383 struct tracepoint_entry *entry; 384 void *old; 385 int ret = -ENOENT; 386 387 mutex_lock(&tracepoints_mutex); 388 entry = get_tracepoint(name); 389 if (!entry) 390 goto end; 391 if (entry->rcu_pending) 392 rcu_barrier_sched(); 393 old = tracepoint_entry_remove_probe(entry, probe); 394 if (!old) { 395 printk(KERN_WARNING "Warning: Trying to unregister a probe" 396 "that doesn't exist\n"); 397 goto end; 398 } 399 mutex_unlock(&tracepoints_mutex); 400 tracepoint_update_probes(); /* may update entry */ 401 mutex_lock(&tracepoints_mutex); 402 entry = get_tracepoint(name); 403 if (!entry) 404 goto end; 405 if (entry->rcu_pending) 406 rcu_barrier_sched(); 407 tracepoint_entry_free_old(entry, old); 408 remove_tracepoint(name); /* Ignore busy error message */ 409 ret = 0; 410 end: 411 mutex_unlock(&tracepoints_mutex); 412 return ret; 413 } 414 EXPORT_SYMBOL_GPL(tracepoint_probe_unregister); 415 416 /** 417 * tracepoint_get_iter_range - Get a next tracepoint iterator given a range. 418 * @tracepoint: current tracepoints (in), next tracepoint (out) 419 * @begin: beginning of the range 420 * @end: end of the range 421 * 422 * Returns whether a next tracepoint has been found (1) or not (0). 423 * Will return the first tracepoint in the range if the input tracepoint is 424 * NULL. 425 */ 426 int tracepoint_get_iter_range(struct tracepoint **tracepoint, 427 struct tracepoint *begin, struct tracepoint *end) 428 { 429 if (!*tracepoint && begin != end) { 430 *tracepoint = begin; 431 return 1; 432 } 433 if (*tracepoint >= begin && *tracepoint < end) 434 return 1; 435 return 0; 436 } 437 EXPORT_SYMBOL_GPL(tracepoint_get_iter_range); 438 439 static void tracepoint_get_iter(struct tracepoint_iter *iter) 440 { 441 int found = 0; 442 443 /* Core kernel tracepoints */ 444 if (!iter->module) { 445 found = tracepoint_get_iter_range(&iter->tracepoint, 446 __start___tracepoints, __stop___tracepoints); 447 if (found) 448 goto end; 449 } 450 /* tracepoints in modules. */ 451 found = module_get_iter_tracepoints(iter); 452 end: 453 if (!found) 454 tracepoint_iter_reset(iter); 455 } 456 457 void tracepoint_iter_start(struct tracepoint_iter *iter) 458 { 459 tracepoint_get_iter(iter); 460 } 461 EXPORT_SYMBOL_GPL(tracepoint_iter_start); 462 463 void tracepoint_iter_next(struct tracepoint_iter *iter) 464 { 465 iter->tracepoint++; 466 /* 467 * iter->tracepoint may be invalid because we blindly incremented it. 468 * Make sure it is valid by marshalling on the tracepoints, getting the 469 * tracepoints from following modules if necessary. 470 */ 471 tracepoint_get_iter(iter); 472 } 473 EXPORT_SYMBOL_GPL(tracepoint_iter_next); 474 475 void tracepoint_iter_stop(struct tracepoint_iter *iter) 476 { 477 } 478 EXPORT_SYMBOL_GPL(tracepoint_iter_stop); 479 480 void tracepoint_iter_reset(struct tracepoint_iter *iter) 481 { 482 iter->module = NULL; 483 iter->tracepoint = NULL; 484 } 485 EXPORT_SYMBOL_GPL(tracepoint_iter_reset); 486