Merge tag 'trace-3.15-v2' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt...
authorLinus Torvalds <torvalds@linux-foundation.org>
Sat, 12 Apr 2014 20:06:10 +0000 (13:06 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Sat, 12 Apr 2014 20:06:10 +0000 (13:06 -0700)
Pull more tracing updates from Steven Rostedt:
 "This includes the final patch to clean up and fix the issue with the
  design of tracepoints and how a user could register a tracepoint and
  have that tracepoint not be activated but no error was shown.

  The design was for an out of tree module but broke in tree users.  The
  clean up was to remove the saving of the hash table of tracepoint
  names such that they can be enabled before they exist (enabling a
  module tracepoint before that module is loaded).  This added more
  complexity than needed.  The clean up was to remove that code and just
  enable tracepoints that exist or fail if they do not.

  This removed a lot of code as well as the complexity that it brought.
  As a side effect, instead of registering a tracepoint by its name, the
  tracepoint needs to be registered with the tracepoint descriptor.
  This removes having to duplicate the tracepoint names that are
  enabled.

  The second patch was added that simplified the way modules were
  searched for.

  This cleanup required changes that were in the 3.15 queue as well as
  some changes that were added late in the 3.14-rc cycle.  This final
  change waited till the two were merged in upstream and then the change
  was added and full tests were run.  Unfortunately, the test found some
  errors, but after it was already submitted to the for-next branch and
  not to be rebased.  Sparse errors were detected by Fengguang Wu's bot
  tests, and my internal tests discovered that the anonymous union
  initialization triggered a bug in older gcc compilers.  Luckily, there
  was a bugzilla for the gcc bug which gave a work around to the
  problem.  The third and fourth patch handled the sparse error and the
  gcc bug respectively.

  A final patch was tagged along to fix a missing documentation for the
  README file"

* tag 'trace-3.15-v2' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace:
  tracing: Add missing function triggers dump and cpudump to README
  tracing: Fix anonymous unions in struct ftrace_event_call
  tracepoint: Fix sparse warnings in tracepoint.c
  tracepoint: Simplify tracepoint module search
  tracepoint: Use struct pointer instead of name hash for reg/unreg tracepoints

1  2 
include/linux/syscalls.h
kernel/tracepoint.c

diff --combined include/linux/syscalls.h
index 697ceb70a9a9107689c1fc98f8599552314a9fd8,af94c98087c46cd4a4cab2f6901c36f3b39646a1..a4a0588c5397f5e0057a3565a45a60ed3d92c4d9
@@@ -119,8 -119,10 +119,10 @@@ extern struct trace_event_functions exi
        static struct syscall_metadata __syscall_meta_##sname;          \
        static struct ftrace_event_call __used                          \
          event_enter_##sname = {                                       \
-               .name                   = "sys_enter"#sname,            \
                .class                  = &event_class_syscall_enter,   \
+               {                                                       \
+                       .name                   = "sys_enter"#sname,    \
+               },                                                      \
                .event.funcs            = &enter_syscall_print_funcs,   \
                .data                   = (void *)&__syscall_meta_##sname,\
                .flags                  = TRACE_EVENT_FL_CAP_ANY,       \
        static struct syscall_metadata __syscall_meta_##sname;          \
        static struct ftrace_event_call __used                          \
          event_exit_##sname = {                                        \
-               .name                   = "sys_exit"#sname,             \
                .class                  = &event_class_syscall_exit,    \
+               {                                                       \
+                       .name                   = "sys_exit"#sname,     \
+               },                                                      \
                .event.funcs            = &exit_syscall_print_funcs,    \
                .data                   = (void *)&__syscall_meta_##sname,\
                .flags                  = TRACE_EVENT_FL_CAP_ANY,       \
        }                                                               \
        static inline long SYSC##name(__MAP(x,__SC_DECL,__VA_ARGS__))
  
 +asmlinkage long sys32_quotactl(unsigned int cmd, const char __user *special,
 +                             qid_t id, void __user *addr);
  asmlinkage long sys_time(time_t __user *tloc);
  asmlinkage long sys_stime(time_t __user *tptr);
  asmlinkage long sys_gettimeofday(struct timeval __user *tv,
@@@ -748,9 -750,6 +752,9 @@@ asmlinkage long sys_linkat(int olddfd, 
                           int newdfd, const char __user *newname, int flags);
  asmlinkage long sys_renameat(int olddfd, const char __user * oldname,
                             int newdfd, const char __user * newname);
 +asmlinkage long sys_renameat2(int olddfd, const char __user *oldname,
 +                            int newdfd, const char __user *newname,
 +                            unsigned int flags);
  asmlinkage long sys_futimesat(int dfd, const char __user *filename,
                              struct timeval __user *utimes);
  asmlinkage long sys_faccessat(int dfd, const char __user *filename, int mode);
diff --combined kernel/tracepoint.c
index fb0a38a265555c6c846254859c83468d0aba0f35,ca2cfe21bb8e22ab02b3cb7a5c3536dcc8309a6f..ac5b23cf7212c6ebb0045bedce2f13ee8d0a19b3
@@@ -1,5 -1,5 +1,5 @@@
  /*
-  * Copyright (C) 2008 Mathieu Desnoyers
+  * Copyright (C) 2008-2014 Mathieu Desnoyers
   *
   * This program is free software; you can redistribute it and/or modify
   * it under the terms of the GNU General Public License as published by
@@@ -33,39 -33,27 +33,27 @@@ extern struct tracepoint * const __stop
  /* Set to 1 to enable tracepoint debug output */
  static const int tracepoint_debug;
  
+ #ifdef CONFIG_MODULES
  /*
-  * Tracepoints mutex protects the builtin and module tracepoints and the hash
-  * table, as well as the local module list.
+  * Tracepoint module list mutex protects the local module list.
   */
- static DEFINE_MUTEX(tracepoints_mutex);
+ static DEFINE_MUTEX(tracepoint_module_list_mutex);
  
- #ifdef CONFIG_MODULES
- /* Local list of struct module */
+ /* Local list of struct tp_module */
  static LIST_HEAD(tracepoint_module_list);
  #endif /* CONFIG_MODULES */
  
  /*
-  * Tracepoint hash table, containing the active tracepoints.
-  * Protected by tracepoints_mutex.
+  * tracepoints_mutex protects the builtin and module tracepoints.
+  * tracepoints_mutex nests inside tracepoint_module_list_mutex.
   */
- #define TRACEPOINT_HASH_BITS 6
- #define TRACEPOINT_TABLE_SIZE (1 << TRACEPOINT_HASH_BITS)
- static struct hlist_head tracepoint_table[TRACEPOINT_TABLE_SIZE];
+ static DEFINE_MUTEX(tracepoints_mutex);
  
  /*
   * Note about RCU :
   * It is used to delay the free of multiple probes array until a quiescent
   * state is reached.
-  * Tracepoint entries modifications are protected by the tracepoints_mutex.
   */
- struct tracepoint_entry {
-       struct hlist_node hlist;
-       struct tracepoint_func *funcs;
-       int refcount;   /* Number of times armed. 0 if disarmed. */
-       int enabled;    /* Tracepoint enabled */
-       char name[0];
- };
  struct tp_probes {
        struct rcu_head rcu;
        struct tracepoint_func probes[0];
@@@ -92,34 -80,33 +80,33 @@@ static inline void release_probes(struc
        }
  }
  
- static void debug_print_probes(struct tracepoint_entry *entry)
+ static void debug_print_probes(struct tracepoint_func *funcs)
  {
        int i;
  
-       if (!tracepoint_debug || !entry->funcs)
+       if (!tracepoint_debug || !funcs)
                return;
  
-       for (i = 0; entry->funcs[i].func; i++)
-               printk(KERN_DEBUG "Probe %d : %p\n", i, entry->funcs[i].func);
+       for (i = 0; funcs[i].func; i++)
+               printk(KERN_DEBUG "Probe %d : %p\n", i, funcs[i].func);
  }
  
- static struct tracepoint_func *
- tracepoint_entry_add_probe(struct tracepoint_entry *entry,
-                          void *probe, void *data)
+ static struct tracepoint_func *func_add(struct tracepoint_func **funcs,
+               struct tracepoint_func *tp_func)
  {
        int nr_probes = 0;
        struct tracepoint_func *old, *new;
  
-       if (WARN_ON(!probe))
+       if (WARN_ON(!tp_func->func))
                return ERR_PTR(-EINVAL);
  
-       debug_print_probes(entry);
-       old = entry->funcs;
+       debug_print_probes(*funcs);
+       old = *funcs;
        if (old) {
                /* (N -> N+1), (N != 0, 1) probes */
                for (nr_probes = 0; old[nr_probes].func; nr_probes++)
-                       if (old[nr_probes].func == probe &&
-                           old[nr_probes].data == data)
+                       if (old[nr_probes].func == tp_func->func &&
+                           old[nr_probes].data == tp_func->data)
                                return ERR_PTR(-EEXIST);
        }
        /* + 2 : one for new probe, one for NULL func */
                return ERR_PTR(-ENOMEM);
        if (old)
                memcpy(new, old, nr_probes * sizeof(struct tracepoint_func));
-       new[nr_probes].func = probe;
-       new[nr_probes].data = data;
+       new[nr_probes] = *tp_func;
        new[nr_probes + 1].func = NULL;
-       entry->refcount = nr_probes + 1;
-       entry->funcs = new;
-       debug_print_probes(entry);
+       *funcs = new;
+       debug_print_probes(*funcs);
        return old;
  }
  
- static void *
- tracepoint_entry_remove_probe(struct tracepoint_entry *entry,
-                             void *probe, void *data)
+ static void *func_remove(struct tracepoint_func **funcs,
+               struct tracepoint_func *tp_func)
  {
        int nr_probes = 0, nr_del = 0, i;
        struct tracepoint_func *old, *new;
  
-       old = entry->funcs;
+       old = *funcs;
  
        if (!old)
                return ERR_PTR(-ENOENT);
  
-       debug_print_probes(entry);
+       debug_print_probes(*funcs);
        /* (N -> M), (N > 1, M >= 0) probes */
-       if (probe) {
+       if (tp_func->func) {
                for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
-                       if (old[nr_probes].func == probe &&
-                            old[nr_probes].data == data)
+                       if (old[nr_probes].func == tp_func->func &&
+                            old[nr_probes].data == tp_func->data)
                                nr_del++;
                }
        }
         */
        if (nr_probes - nr_del == 0) {
                /* N -> 0, (N > 1) */
-               entry->funcs = NULL;
-               entry->refcount = 0;
-               debug_print_probes(entry);
+               *funcs = NULL;
+               debug_print_probes(*funcs);
                return old;
        } else {
                int j = 0;
                if (new == NULL)
                        return ERR_PTR(-ENOMEM);
                for (i = 0; old[i].func; i++)
-                       if (old[i].func != probe || old[i].data != data)
+                       if (old[i].func != tp_func->func
+                                       || old[i].data != tp_func->data)
                                new[j++] = old[i];
                new[nr_probes - nr_del].func = NULL;
-               entry->refcount = nr_probes - nr_del;
-               entry->funcs = new;
+               *funcs = new;
        }
-       debug_print_probes(entry);
+       debug_print_probes(*funcs);
        return old;
  }
  
  /*
-  * Get tracepoint if the tracepoint is present in the tracepoint hash table.
-  * Must be called with tracepoints_mutex held.
-  * Returns NULL if not present.
+  * Add the probe function to a tracepoint.
   */
- static struct tracepoint_entry *get_tracepoint(const char *name)
+ static int tracepoint_add_func(struct tracepoint *tp,
+               struct tracepoint_func *func)
  {
-       struct hlist_head *head;
-       struct tracepoint_entry *e;
-       u32 hash = jhash(name, strlen(name), 0);
-       head = &tracepoint_table[hash & (TRACEPOINT_TABLE_SIZE - 1)];
-       hlist_for_each_entry(e, head, hlist) {
-               if (!strcmp(name, e->name))
-                       return e;
-       }
-       return NULL;
- }
+       struct tracepoint_func *old, *tp_funcs;
  
- /*
-  * Add the tracepoint to the tracepoint hash table. Must be called with
-  * tracepoints_mutex held.
-  */
- static struct tracepoint_entry *add_tracepoint(const char *name)
- {
-       struct hlist_head *head;
-       struct tracepoint_entry *e;
-       size_t name_len = strlen(name) + 1;
-       u32 hash = jhash(name, name_len-1, 0);
-       head = &tracepoint_table[hash & (TRACEPOINT_TABLE_SIZE - 1)];
-       hlist_for_each_entry(e, head, hlist) {
-               if (!strcmp(name, e->name)) {
-                       printk(KERN_NOTICE
-                               "tracepoint %s busy\n", name);
-                       return ERR_PTR(-EEXIST);        /* Already there */
-               }
-       }
-       /*
-        * Using kmalloc here to allocate a variable length element. Could
-        * cause some memory fragmentation if overused.
-        */
-       e = kmalloc(sizeof(struct tracepoint_entry) + name_len, GFP_KERNEL);
-       if (!e)
-               return ERR_PTR(-ENOMEM);
-       memcpy(&e->name[0], name, name_len);
-       e->funcs = NULL;
-       e->refcount = 0;
-       e->enabled = 0;
-       hlist_add_head(&e->hlist, head);
-       return e;
- }
+       if (tp->regfunc && !static_key_enabled(&tp->key))
+               tp->regfunc();
  
- /*
-  * Remove the tracepoint from the tracepoint hash table. Must be called with
-  * mutex_lock held.
-  */
- static inline void remove_tracepoint(struct tracepoint_entry *e)
- {
-       hlist_del(&e->hlist);
-       kfree(e);
- }
- /*
-  * Sets the probe callback corresponding to one tracepoint.
-  */
- static void set_tracepoint(struct tracepoint_entry **entry,
-       struct tracepoint *elem, int active)
- {
-       WARN_ON(strcmp((*entry)->name, elem->name) != 0);
-       if (elem->regfunc && !static_key_enabled(&elem->key) && active)
-               elem->regfunc();
-       else if (elem->unregfunc && static_key_enabled(&elem->key) && !active)
-               elem->unregfunc();
+       tp_funcs = rcu_dereference_protected(tp->funcs,
+                       lockdep_is_held(&tracepoints_mutex));
+       old = func_add(&tp_funcs, func);
+       if (IS_ERR(old)) {
+               WARN_ON_ONCE(1);
+               return PTR_ERR(old);
+       }
+       release_probes(old);
  
        /*
         * rcu_assign_pointer has a smp_wmb() which makes sure that the new
         * include/linux/tracepoints.h. A matching smp_read_barrier_depends()
         * is used.
         */
-       rcu_assign_pointer(elem->funcs, (*entry)->funcs);
-       if (active && !static_key_enabled(&elem->key))
-               static_key_slow_inc(&elem->key);
-       else if (!active && static_key_enabled(&elem->key))
-               static_key_slow_dec(&elem->key);
+       rcu_assign_pointer(tp->funcs, tp_funcs);
+       if (!static_key_enabled(&tp->key))
+               static_key_slow_inc(&tp->key);
+       return 0;
  }
  
  /*
-  * Disable a tracepoint and its probe callback.
+  * Remove a probe function from a tracepoint.
   * Note: only waiting an RCU period after setting elem->call to the empty
   * function insures that the original callback is not used anymore. This insured
   * by preempt_disable around the call site.
   */
- static void disable_tracepoint(struct tracepoint *elem)
+ static int tracepoint_remove_func(struct tracepoint *tp,
+               struct tracepoint_func *func)
  {
-       if (elem->unregfunc && static_key_enabled(&elem->key))
-               elem->unregfunc();
-       if (static_key_enabled(&elem->key))
-               static_key_slow_dec(&elem->key);
-       rcu_assign_pointer(elem->funcs, NULL);
- }
+       struct tracepoint_func *old, *tp_funcs;
  
- /**
-  * tracepoint_update_probe_range - Update a probe range
-  * @begin: beginning of the range
-  * @end: end of the range
-  *
-  * Updates the probe callback corresponding to a range of tracepoints.
-  * Called with tracepoints_mutex held.
-  */
- static void tracepoint_update_probe_range(struct tracepoint * const *begin,
-                                         struct tracepoint * const *end)
- {
-       struct tracepoint * const *iter;
-       struct tracepoint_entry *mark_entry;
-       if (!begin)
-               return;
-       for (iter = begin; iter < end; iter++) {
-               mark_entry = get_tracepoint((*iter)->name);
-               if (mark_entry) {
-                       set_tracepoint(&mark_entry, *iter,
-                                       !!mark_entry->refcount);
-                       mark_entry->enabled = !!mark_entry->refcount;
-               } else {
-                       disable_tracepoint(*iter);
-               }
+       tp_funcs = rcu_dereference_protected(tp->funcs,
+                       lockdep_is_held(&tracepoints_mutex));
+       old = func_remove(&tp_funcs, func);
+       if (IS_ERR(old)) {
+               WARN_ON_ONCE(1);
+               return PTR_ERR(old);
        }
- }
- #ifdef CONFIG_MODULES
- void module_update_tracepoints(void)
- {
-       struct tp_module *tp_mod;
-       list_for_each_entry(tp_mod, &tracepoint_module_list, list)
-               tracepoint_update_probe_range(tp_mod->tracepoints_ptrs,
-                       tp_mod->tracepoints_ptrs + tp_mod->num_tracepoints);
- }
- #else /* CONFIG_MODULES */
- void module_update_tracepoints(void)
- {
- }
- #endif /* CONFIG_MODULES */
+       release_probes(old);
  
+       if (!tp_funcs) {
+               /* Removed last function */
+               if (tp->unregfunc && static_key_enabled(&tp->key))
+                       tp->unregfunc();
  
- /*
-  * Update probes, removing the faulty probes.
-  * Called with tracepoints_mutex held.
-  */
- static void tracepoint_update_probes(void)
- {
-       /* Core kernel tracepoints */
-       tracepoint_update_probe_range(__start___tracepoints_ptrs,
-               __stop___tracepoints_ptrs);
-       /* tracepoints in modules. */
-       module_update_tracepoints();
- }
- static struct tracepoint_func *
- tracepoint_add_probe(const char *name, void *probe, void *data)
- {
-       struct tracepoint_entry *entry;
-       struct tracepoint_func *old;
-       entry = get_tracepoint(name);
-       if (!entry) {
-               entry = add_tracepoint(name);
-               if (IS_ERR(entry))
-                       return (struct tracepoint_func *)entry;
+               if (static_key_enabled(&tp->key))
+                       static_key_slow_dec(&tp->key);
        }
-       old = tracepoint_entry_add_probe(entry, probe, data);
-       if (IS_ERR(old) && !entry->refcount)
-               remove_tracepoint(entry);
-       return old;
+       rcu_assign_pointer(tp->funcs, tp_funcs);
+       return 0;
  }
  
  /**
   * tracepoint_probe_register -  Connect a probe to a tracepoint
-  * @name: tracepoint name
+  * @tp: tracepoint
   * @probe: probe handler
-  * @data: probe private data
-  *
-  * Returns:
-  * - 0 if the probe was successfully registered, and tracepoint
-  *   callsites are currently loaded for that probe,
-  * - -ENODEV if the probe was successfully registered, but no tracepoint
-  *   callsite is currently loaded for that probe,
-  * - other negative error value on error.
-  *
-  * When tracepoint_probe_register() returns either 0 or -ENODEV,
-  * parameters @name, @probe, and @data may be used by the tracepoint
-  * infrastructure until the probe is unregistered.
   *
-  * The probe address must at least be aligned on the architecture pointer size.
+  * Returns 0 if ok, error value on error.
+  * Note: if @tp is within a module, the caller is responsible for
+  * unregistering the probe before the module is gone. This can be
+  * performed either with a tracepoint module going notifier, or from
+  * within module exit functions.
   */
- int tracepoint_probe_register(const char *name, void *probe, void *data)
+ int tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data)
  {
-       struct tracepoint_func *old;
-       struct tracepoint_entry *entry;
-       int ret = 0;
+       struct tracepoint_func tp_func;
+       int ret;
  
        mutex_lock(&tracepoints_mutex);
-       old = tracepoint_add_probe(name, probe, data);
-       if (IS_ERR(old)) {
-               mutex_unlock(&tracepoints_mutex);
-               return PTR_ERR(old);
-       }
-       tracepoint_update_probes();             /* may update entry */
-       entry = get_tracepoint(name);
-       /* Make sure the entry was enabled */
-       if (!entry || !entry->enabled)
-               ret = -ENODEV;
+       tp_func.func = probe;
+       tp_func.data = data;
+       ret = tracepoint_add_func(tp, &tp_func);
        mutex_unlock(&tracepoints_mutex);
-       release_probes(old);
        return ret;
  }
  EXPORT_SYMBOL_GPL(tracepoint_probe_register);
  
- static struct tracepoint_func *
- tracepoint_remove_probe(const char *name, void *probe, void *data)
- {
-       struct tracepoint_entry *entry;
-       struct tracepoint_func *old;
-       entry = get_tracepoint(name);
-       if (!entry)
-               return ERR_PTR(-ENOENT);
-       old = tracepoint_entry_remove_probe(entry, probe, data);
-       if (IS_ERR(old))
-               return old;
-       if (!entry->refcount)
-               remove_tracepoint(entry);
-       return old;
- }
  /**
   * tracepoint_probe_unregister -  Disconnect a probe from a tracepoint
-  * @name: tracepoint name
+  * @tp: tracepoint
   * @probe: probe function pointer
-  * @data: probe private data
   *
-  * We do not need to call a synchronize_sched to make sure the probes have
-  * finished running before doing a module unload, because the module unload
-  * itself uses stop_machine(), which insures that every preempt disabled section
-  * have finished.
+  * Returns 0 if ok, error value on error.
   */
- int tracepoint_probe_unregister(const char *name, void *probe, void *data)
+ int tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data)
  {
-       struct tracepoint_func *old;
+       struct tracepoint_func tp_func;
+       int ret;
  
        mutex_lock(&tracepoints_mutex);
-       old = tracepoint_remove_probe(name, probe, data);
-       if (IS_ERR(old)) {
-               mutex_unlock(&tracepoints_mutex);
-               return PTR_ERR(old);
-       }
-       tracepoint_update_probes();             /* may update entry */
+       tp_func.func = probe;
+       tp_func.data = data;
+       ret = tracepoint_remove_func(tp, &tp_func);
        mutex_unlock(&tracepoints_mutex);
-       release_probes(old);
-       return 0;
+       return ret;
  }
  EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
  
  #ifdef CONFIG_MODULES
  bool trace_module_has_bad_taint(struct module *mod)
  {
 -      return mod->taints & ~((1 << TAINT_OOT_MODULE) | (1 << TAINT_CRAP));
 +      return mod->taints & ~((1 << TAINT_OOT_MODULE) | (1 << TAINT_CRAP) |
 +                             (1 << TAINT_UNSIGNED_MODULE));
  }
  
+ static BLOCKING_NOTIFIER_HEAD(tracepoint_notify_list);
+ /**
+  * register_tracepoint_notifier - register tracepoint coming/going notifier
+  * @nb: notifier block
+  *
+  * Notifiers registered with this function are called on module
+  * coming/going with the tracepoint_module_list_mutex held.
+  * The notifier block callback should expect a "struct tp_module" data
+  * pointer.
+  */
+ int register_tracepoint_module_notifier(struct notifier_block *nb)
+ {
+       struct tp_module *tp_mod;
+       int ret;
+       mutex_lock(&tracepoint_module_list_mutex);
+       ret = blocking_notifier_chain_register(&tracepoint_notify_list, nb);
+       if (ret)
+               goto end;
+       list_for_each_entry(tp_mod, &tracepoint_module_list, list)
+               (void) nb->notifier_call(nb, MODULE_STATE_COMING, tp_mod);
+ end:
+       mutex_unlock(&tracepoint_module_list_mutex);
+       return ret;
+ }
+ EXPORT_SYMBOL_GPL(register_tracepoint_module_notifier);
+ /**
+  * unregister_tracepoint_notifier - unregister tracepoint coming/going notifier
+  * @nb: notifier block
+  *
+  * The notifier block callback should expect a "struct tp_module" data
+  * pointer.
+  */
+ int unregister_tracepoint_module_notifier(struct notifier_block *nb)
+ {
+       struct tp_module *tp_mod;
+       int ret;
+       mutex_lock(&tracepoint_module_list_mutex);
+       ret = blocking_notifier_chain_unregister(&tracepoint_notify_list, nb);
+       if (ret)
+               goto end;
+       list_for_each_entry(tp_mod, &tracepoint_module_list, list)
+               (void) nb->notifier_call(nb, MODULE_STATE_GOING, tp_mod);
+ end:
+       mutex_unlock(&tracepoint_module_list_mutex);
+       return ret;
+ }
+ EXPORT_SYMBOL_GPL(unregister_tracepoint_module_notifier);
+ /*
+  * Ensure the tracer unregistered the module's probes before the module
+  * teardown is performed. Prevents leaks of probe and data pointers.
+  */
+ static void tp_module_going_check_quiescent(struct tracepoint * const *begin,
+               struct tracepoint * const *end)
+ {
+       struct tracepoint * const *iter;
+       if (!begin)
+               return;
+       for (iter = begin; iter < end; iter++)
+               WARN_ON_ONCE((*iter)->funcs);
+ }
  static int tracepoint_module_coming(struct module *mod)
  {
        struct tp_module *tp_mod;
        /*
         * We skip modules that taint the kernel, especially those with different
         * module headers (for forced load), to make sure we don't cause a crash.
 -       * Staging and out-of-tree GPL modules are fine.
 +       * Staging, out-of-tree, and unsigned GPL modules are fine.
         */
        if (trace_module_has_bad_taint(mod))
                return 0;
-       mutex_lock(&tracepoints_mutex);
+       mutex_lock(&tracepoint_module_list_mutex);
        tp_mod = kmalloc(sizeof(struct tp_module), GFP_KERNEL);
        if (!tp_mod) {
                ret = -ENOMEM;
                goto end;
        }
-       tp_mod->num_tracepoints = mod->num_tracepoints;
-       tp_mod->tracepoints_ptrs = mod->tracepoints_ptrs;
+       tp_mod->mod = mod;
        list_add_tail(&tp_mod->list, &tracepoint_module_list);
-       tracepoint_update_probe_range(mod->tracepoints_ptrs,
-               mod->tracepoints_ptrs + mod->num_tracepoints);
+       blocking_notifier_call_chain(&tracepoint_notify_list,
+                       MODULE_STATE_COMING, tp_mod);
  end:
-       mutex_unlock(&tracepoints_mutex);
+       mutex_unlock(&tracepoint_module_list_mutex);
        return ret;
  }
  
- static int tracepoint_module_going(struct module *mod)
+ static void tracepoint_module_going(struct module *mod)
  {
-       struct tp_module *pos;
+       struct tp_module *tp_mod;
  
        if (!mod->num_tracepoints)
-               return 0;
+               return;
  
-       mutex_lock(&tracepoints_mutex);
-       tracepoint_update_probe_range(mod->tracepoints_ptrs,
-               mod->tracepoints_ptrs + mod->num_tracepoints);
-       list_for_each_entry(pos, &tracepoint_module_list, list) {
-               if (pos->tracepoints_ptrs == mod->tracepoints_ptrs) {
-                       list_del(&pos->list);
-                       kfree(pos);
+       mutex_lock(&tracepoint_module_list_mutex);
+       list_for_each_entry(tp_mod, &tracepoint_module_list, list) {
+               if (tp_mod->mod == mod) {
+                       blocking_notifier_call_chain(&tracepoint_notify_list,
+                                       MODULE_STATE_GOING, tp_mod);
+                       list_del(&tp_mod->list);
+                       kfree(tp_mod);
+                       /*
+                        * Called the going notifier before checking for
+                        * quiescence.
+                        */
+                       tp_module_going_check_quiescent(mod->tracepoints_ptrs,
+                               mod->tracepoints_ptrs + mod->num_tracepoints);
                        break;
                }
        }
         * flag on "going", in case a module taints the kernel only after being
         * loaded.
         */
-       mutex_unlock(&tracepoints_mutex);
-       return 0;
+       mutex_unlock(&tracepoint_module_list_mutex);
  }
  
- int tracepoint_module_notify(struct notifier_block *self,
-                            unsigned long val, void *data)
static int tracepoint_module_notify(struct notifier_block *self,
+               unsigned long val, void *data)
  {
        struct module *mod = data;
        int ret = 0;
        case MODULE_STATE_LIVE:
                break;
        case MODULE_STATE_GOING:
-               ret = tracepoint_module_going(mod);
+               tracepoint_module_going(mod);
+               break;
+       case MODULE_STATE_UNFORMED:
                break;
        }
        return ret;
  }
  
- struct notifier_block tracepoint_module_nb = {
+ static struct notifier_block tracepoint_module_nb = {
        .notifier_call = tracepoint_module_notify,
        .priority = 0,
  };
  
- static int init_tracepoints(void)
+ static __init int init_tracepoints(void)
  {
-       return register_module_notifier(&tracepoint_module_nb);
+       int ret;
+       ret = register_module_notifier(&tracepoint_module_nb);
+       if (ret)
+               pr_warning("Failed to register tracepoint module enter notifier\n");
+       return ret;
  }
  __initcall(init_tracepoints);
  #endif /* CONFIG_MODULES */
  
+ static void for_each_tracepoint_range(struct tracepoint * const *begin,
+               struct tracepoint * const *end,
+               void (*fct)(struct tracepoint *tp, void *priv),
+               void *priv)
+ {
+       struct tracepoint * const *iter;
+       if (!begin)
+               return;
+       for (iter = begin; iter < end; iter++)
+               fct(*iter, priv);
+ }
+ /**
+  * for_each_kernel_tracepoint - iteration on all kernel tracepoints
+  * @fct: callback
+  * @priv: private data
+  */
+ void for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
+               void *priv)
+ {
+       for_each_tracepoint_range(__start___tracepoints_ptrs,
+               __stop___tracepoints_ptrs, fct, priv);
+ }
+ EXPORT_SYMBOL_GPL(for_each_kernel_tracepoint);
  #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
  
  /* NB: reg/unreg are called while guarded with the tracepoints_mutex */