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