mac80211: fix deauth before assoc
[safe/jmp/linux-2.6] / kernel / lockdep.c
index 2f09702..ec21304 100644 (file)
 #include <linux/ftrace.h>
 #include <linux/stringify.h>
 #include <linux/bitops.h>
+#include <linux/gfp.h>
 
 #include <asm/sections.h>
 
 #include "lockdep_internals.h"
 
 #define CREATE_TRACE_POINTS
-#include <trace/events/lockdep.h>
+#include <trace/events/lock.h>
 
 #ifdef CONFIG_PROVE_LOCKING
 int prove_locking = 1;
@@ -73,11 +74,11 @@ module_param(lock_stat, int, 0644);
  * to use a raw spinlock - we really dont want the spinlock
  * code to recurse back into the lockdep code...
  */
-static raw_spinlock_t lockdep_lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
+static arch_spinlock_t lockdep_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
 
 static int graph_lock(void)
 {
-       __raw_spin_lock(&lockdep_lock);
+       arch_spin_lock(&lockdep_lock);
        /*
         * Make sure that if another CPU detected a bug while
         * walking the graph we dont change it (while the other
@@ -85,7 +86,7 @@ static int graph_lock(void)
         * dropped already)
         */
        if (!debug_locks) {
-               __raw_spin_unlock(&lockdep_lock);
+               arch_spin_unlock(&lockdep_lock);
                return 0;
        }
        /* prevent any recursions within lockdep from causing deadlocks */
@@ -95,11 +96,11 @@ static int graph_lock(void)
 
 static inline int graph_unlock(void)
 {
-       if (debug_locks && !__raw_spin_is_locked(&lockdep_lock))
+       if (debug_locks && !arch_spin_is_locked(&lockdep_lock))
                return DEBUG_LOCKS_WARN_ON(1);
 
        current->lockdep_recursion--;
-       __raw_spin_unlock(&lockdep_lock);
+       arch_spin_unlock(&lockdep_lock);
        return 0;
 }
 
@@ -111,7 +112,7 @@ static inline int debug_locks_off_graph_unlock(void)
 {
        int ret = debug_locks_off();
 
-       __raw_spin_unlock(&lockdep_lock);
+       arch_spin_unlock(&lockdep_lock);
 
        return ret;
 }
@@ -140,7 +141,13 @@ static inline struct lock_class *hlock_class(struct held_lock *hlock)
 }
 
 #ifdef CONFIG_LOCK_STAT
-static DEFINE_PER_CPU(struct lock_class_stats[MAX_LOCKDEP_KEYS], lock_stats);
+static DEFINE_PER_CPU(struct lock_class_stats[MAX_LOCKDEP_KEYS],
+                     cpu_lock_stats);
+
+static inline u64 lockstat_clock(void)
+{
+       return cpu_clock(smp_processor_id());
+}
 
 static int lock_point(unsigned long points[], unsigned long ip)
 {
@@ -158,12 +165,12 @@ static int lock_point(unsigned long points[], unsigned long ip)
        return i;
 }
 
-static void lock_time_inc(struct lock_time *lt, s64 time)
+static void lock_time_inc(struct lock_time *lt, u64 time)
 {
        if (time > lt->max)
                lt->max = time;
 
-       if (time < lt->min || !lt->min)
+       if (time < lt->min || !lt->nr)
                lt->min = time;
 
        lt->total += time;
@@ -172,8 +179,15 @@ static void lock_time_inc(struct lock_time *lt, s64 time)
 
 static inline void lock_time_add(struct lock_time *src, struct lock_time *dst)
 {
-       dst->min += src->min;
-       dst->max += src->max;
+       if (!src->nr)
+               return;
+
+       if (src->max > dst->max)
+               dst->max = src->max;
+
+       if (src->min < dst->min || !dst->nr)
+               dst->min = src->min;
+
        dst->total += src->total;
        dst->nr += src->nr;
 }
@@ -186,7 +200,7 @@ struct lock_class_stats lock_stats(struct lock_class *class)
        memset(&stats, 0, sizeof(struct lock_class_stats));
        for_each_possible_cpu(cpu) {
                struct lock_class_stats *pcs =
-                       &per_cpu(lock_stats, cpu)[class - lock_classes];
+                       &per_cpu(cpu_lock_stats, cpu)[class - lock_classes];
 
                for (i = 0; i < ARRAY_SIZE(stats.contention_point); i++)
                        stats.contention_point[i] += pcs->contention_point[i];
@@ -213,7 +227,7 @@ void clear_lock_stats(struct lock_class *class)
 
        for_each_possible_cpu(cpu) {
                struct lock_class_stats *cpu_stats =
-                       &per_cpu(lock_stats, cpu)[class - lock_classes];
+                       &per_cpu(cpu_lock_stats, cpu)[class - lock_classes];
 
                memset(cpu_stats, 0, sizeof(struct lock_class_stats));
        }
@@ -223,23 +237,23 @@ void clear_lock_stats(struct lock_class *class)
 
 static struct lock_class_stats *get_lock_stats(struct lock_class *class)
 {
-       return &get_cpu_var(lock_stats)[class - lock_classes];
+       return &get_cpu_var(cpu_lock_stats)[class - lock_classes];
 }
 
 static void put_lock_stats(struct lock_class_stats *stats)
 {
-       put_cpu_var(lock_stats);
+       put_cpu_var(cpu_lock_stats);
 }
 
 static void lock_release_holdtime(struct held_lock *hlock)
 {
        struct lock_class_stats *stats;
-       s64 holdtime;
+       u64 holdtime;
 
        if (!lock_stat)
                return;
 
-       holdtime = sched_clock() - hlock->holdtime_stamp;
+       holdtime = lockstat_clock() - hlock->holdtime_stamp;
 
        stats = get_lock_stats(hlock_class(hlock));
        if (hlock->read)
@@ -374,7 +388,8 @@ static int save_trace(struct stack_trace *trace)
         * complete trace that maxes out the entries provided will be reported
         * as incomplete, friggin useless </rant>
         */
-       if (trace->entries[trace->nr_entries-1] == ULONG_MAX)
+       if (trace->nr_entries != 0 &&
+           trace->entries[trace->nr_entries-1] == ULONG_MAX)
                trace->nr_entries--;
 
        trace->max_entries = trace->nr_entries;
@@ -399,7 +414,6 @@ unsigned int nr_hardirq_chains;
 unsigned int nr_softirq_chains;
 unsigned int nr_process_chains;
 unsigned int max_lockdep_depth;
-unsigned int max_recursion_depth;
 
 #ifdef CONFIG_DEBUG_LOCKDEP
 /*
@@ -417,23 +431,7 @@ static struct stack_trace lockdep_init_trace = {
 /*
  * Various lockdep statistics:
  */
-atomic_t chain_lookup_hits;
-atomic_t chain_lookup_misses;
-atomic_t hardirqs_on_events;
-atomic_t hardirqs_off_events;
-atomic_t redundant_hardirqs_on;
-atomic_t redundant_hardirqs_off;
-atomic_t softirqs_on_events;
-atomic_t softirqs_off_events;
-atomic_t redundant_softirqs_on;
-atomic_t redundant_softirqs_off;
-atomic_t nr_unused_locks;
-atomic_t nr_cyclic_checks;
-atomic_t nr_cyclic_check_recursions;
-atomic_t nr_find_usage_forwards_checks;
-atomic_t nr_find_usage_forwards_recursions;
-atomic_t nr_find_usage_backwards_checks;
-atomic_t nr_find_usage_backwards_recursions;
+DEFINE_PER_CPU(struct lockdep_stats, lockdep_stats);
 #endif
 
 /*
@@ -572,9 +570,6 @@ static int static_obj(void *obj)
        unsigned long start = (unsigned long) &_stext,
                      end   = (unsigned long) &_end,
                      addr  = (unsigned long) obj;
-#ifdef CONFIG_SMP
-       int i;
-#endif
 
        /*
         * static variable?
@@ -582,24 +577,19 @@ static int static_obj(void *obj)
        if ((addr >= start) && (addr < end))
                return 1;
 
-#ifdef CONFIG_SMP
+       if (arch_is_kernel_data(addr))
+               return 1;
+
        /*
-        * percpu var?
+        * in-kernel percpu var?
         */
-       for_each_possible_cpu(i) {
-               start = (unsigned long) &__per_cpu_start + per_cpu_offset(i);
-               end   = (unsigned long) &__per_cpu_start + PERCPU_ENOUGH_ROOM
-                                       + per_cpu_offset(i);
-
-               if ((addr >= start) && (addr < end))
-                       return 1;
-       }
-#endif
+       if (is_kernel_percpu_address(addr))
+               return 1;
 
        /*
-        * module var?
+        * module static or percpu var?
         */
-       return is_module_address(addr);
+       return is_module_address(addr) || is_module_percpu_address(addr);
 }
 
 /*
@@ -745,7 +735,7 @@ register_lock_class(struct lockdep_map *lock, unsigned int subclass, int force)
                return NULL;
        }
        class = lock_classes + nr_lock_classes++;
-       debug_atomic_inc(&nr_unused_locks);
+       debug_atomic_inc(nr_unused_locks);
        class->key = key;
        class->name = lock->name;
        class->subclass = subclass;
@@ -815,7 +805,8 @@ static struct lock_list *alloc_list_entry(void)
  * Add a new dependency to the head of the list:
  */
 static int add_lock_to_list(struct lock_class *class, struct lock_class *this,
-                           struct list_head *head, unsigned long ip, int distance)
+                           struct list_head *head, unsigned long ip,
+                           int distance, struct stack_trace *trace)
 {
        struct lock_list *entry;
        /*
@@ -826,11 +817,9 @@ static int add_lock_to_list(struct lock_class *class, struct lock_class *this,
        if (!entry)
                return 0;
 
-       if (!save_trace(&entry->trace))
-               return 0;
-
        entry->class = this;
        entry->distance = distance;
+       entry->trace = *trace;
        /*
         * Since we never remove from the dependency list, the list can
         * be walked lockless by other CPUs, it's only allocation
@@ -843,29 +832,33 @@ static int add_lock_to_list(struct lock_class *class, struct lock_class *this,
        return 1;
 }
 
-/*For good efficiency of modular, we use power of 2*/
+/*
+ * For good efficiency of modular, we use power of 2
+ */
 #define MAX_CIRCULAR_QUEUE_SIZE                4096UL
 #define CQ_MASK                                (MAX_CIRCULAR_QUEUE_SIZE-1)
 
-/* The circular_queue and helpers is used to implement the
+/*
+ * The circular_queue and helpers is used to implement the
  * breadth-first search(BFS)algorithem, by which we can build
  * the shortest path from the next lock to be acquired to the
  * previous held lock if there is a circular between them.
- * */
+ */
 struct circular_queue {
        unsigned long element[MAX_CIRCULAR_QUEUE_SIZE];
        unsigned int  front, rear;
 };
 
 static struct circular_queue lock_cq;
-static unsigned long bfs_accessed[BITS_TO_LONGS(MAX_LOCKDEP_ENTRIES)];
 
 unsigned int max_bfs_queue_depth;
 
+static unsigned int lockdep_dependency_gen_id;
+
 static inline void __cq_init(struct circular_queue *cq)
 {
        cq->front = cq->rear = 0;
-       bitmap_zero(bfs_accessed, MAX_LOCKDEP_ENTRIES);
+       lockdep_dependency_gen_id++;
 }
 
 static inline int __cq_empty(struct circular_queue *cq)
@@ -907,18 +900,20 @@ static inline void mark_lock_accessed(struct lock_list *lock,
                                        struct lock_list *parent)
 {
        unsigned long nr;
+
        nr = lock - list_entries;
        WARN_ON(nr >= nr_list_entries);
        lock->parent = parent;
-       set_bit(nr, bfs_accessed);
+       lock->class->dep_gen_id = lockdep_dependency_gen_id;
 }
 
 static inline unsigned long lock_accessed(struct lock_list *lock)
 {
        unsigned long nr;
+
        nr = lock - list_entries;
        WARN_ON(nr >= nr_list_entries);
-       return test_bit(nr, bfs_accessed);
+       return lock->class->dep_gen_id == lockdep_dependency_gen_id;
 }
 
 static inline struct lock_list *get_lock_parent(struct lock_list *child)
@@ -1151,9 +1146,9 @@ unsigned long lockdep_count_forward_deps(struct lock_class *class)
        this.class = class;
 
        local_irq_save(flags);
-       __raw_spin_lock(&lockdep_lock);
+       arch_spin_lock(&lockdep_lock);
        ret = __lockdep_count_forward_deps(&this);
-       __raw_spin_unlock(&lockdep_lock);
+       arch_spin_unlock(&lockdep_lock);
        local_irq_restore(flags);
 
        return ret;
@@ -1178,9 +1173,9 @@ unsigned long lockdep_count_backward_deps(struct lock_class *class)
        this.class = class;
 
        local_irq_save(flags);
-       __raw_spin_lock(&lockdep_lock);
+       arch_spin_lock(&lockdep_lock);
        ret = __lockdep_count_backward_deps(&this);
-       __raw_spin_unlock(&lockdep_lock);
+       arch_spin_unlock(&lockdep_lock);
        local_irq_restore(flags);
 
        return ret;
@@ -1196,7 +1191,7 @@ check_noncircular(struct lock_list *root, struct lock_class *target,
 {
        int result;
 
-       debug_atomic_inc(&nr_cyclic_checks);
+       debug_atomic_inc(nr_cyclic_checks);
 
        result = __bfs_forwards(root, target, class_equal, target_entry);
 
@@ -1233,7 +1228,7 @@ find_usage_forwards(struct lock_list *root, enum lock_usage_bit bit,
 {
        int result;
 
-       debug_atomic_inc(&nr_find_usage_forwards_checks);
+       debug_atomic_inc(nr_find_usage_forwards_checks);
 
        result = __bfs_forwards(root, (void *)bit, usage_match, target_entry);
 
@@ -1256,7 +1251,7 @@ find_usage_backwards(struct lock_list *root, enum lock_usage_bit bit,
 {
        int result;
 
-       debug_atomic_inc(&nr_find_usage_backwards_checks);
+       debug_atomic_inc(nr_find_usage_backwards_checks);
 
        result = __bfs_backwards(root, (void *)bit, usage_match, target_entry);
 
@@ -1626,12 +1621,20 @@ check_deadlock(struct task_struct *curr, struct held_lock *next,
  */
 static int
 check_prev_add(struct task_struct *curr, struct held_lock *prev,
-              struct held_lock *next, int distance)
+              struct held_lock *next, int distance, int trylock_loop)
 {
        struct lock_list *entry;
        int ret;
        struct lock_list this;
        struct lock_list *uninitialized_var(target_entry);
+       /*
+        * Static variable, serialized by the graph_lock().
+        *
+        * We use this static variable to save the stack trace in case
+        * we call into this function multiple times due to encountering
+        * trylocks in the held lock stack.
+        */
+       static struct stack_trace trace;
 
        /*
         * Prove that the new <prev> -> <next> dependency would not
@@ -1679,20 +1682,23 @@ check_prev_add(struct task_struct *curr, struct held_lock *prev,
                }
        }
 
+       if (!trylock_loop && !save_trace(&trace))
+               return 0;
+
        /*
         * Ok, all validations passed, add the new lock
         * to the previous lock's dependency list:
         */
        ret = add_lock_to_list(hlock_class(prev), hlock_class(next),
                               &hlock_class(prev)->locks_after,
-                              next->acquire_ip, distance);
+                              next->acquire_ip, distance, &trace);
 
        if (!ret)
                return 0;
 
        ret = add_lock_to_list(hlock_class(next), hlock_class(prev),
                               &hlock_class(next)->locks_before,
-                              next->acquire_ip, distance);
+                              next->acquire_ip, distance, &trace);
        if (!ret)
                return 0;
 
@@ -1722,6 +1728,7 @@ static int
 check_prevs_add(struct task_struct *curr, struct held_lock *next)
 {
        int depth = curr->lockdep_depth;
+       int trylock_loop = 0;
        struct held_lock *hlock;
 
        /*
@@ -1747,7 +1754,8 @@ check_prevs_add(struct task_struct *curr, struct held_lock *next)
                 * added:
                 */
                if (hlock->read != 2) {
-                       if (!check_prev_add(curr, hlock, next, distance))
+                       if (!check_prev_add(curr, hlock, next,
+                                               distance, trylock_loop))
                                return 0;
                        /*
                         * Stop after the first non-trylock entry,
@@ -1770,6 +1778,7 @@ check_prevs_add(struct task_struct *curr, struct held_lock *next)
                if (curr->held_locks[depth].irq_context !=
                                curr->held_locks[depth-1].irq_context)
                        break;
+               trylock_loop = 1;
        }
        return 1;
 out_bug:
@@ -1816,7 +1825,7 @@ static inline int lookup_chain_cache(struct task_struct *curr,
        list_for_each_entry(chain, hash_head, entry) {
                if (chain->chain_key == chain_key) {
 cache_hit:
-                       debug_atomic_inc(&chain_lookup_hits);
+                       debug_atomic_inc(chain_lookup_hits);
                        if (very_verbose(class))
                                printk("\nhash chain already cached, key: "
                                        "%016Lx tail class: [%p] %s\n",
@@ -1881,7 +1890,7 @@ cache_hit:
                chain_hlocks[chain->base + j] = class - lock_classes;
        }
        list_add_tail_rcu(&chain->entry, hash_head);
-       debug_atomic_inc(&chain_lookup_misses);
+       debug_atomic_inc(chain_lookup_misses);
        inc_chains();
 
        return 1;
@@ -2128,7 +2137,7 @@ check_usage_backwards(struct task_struct *curr, struct held_lock *this,
                return ret;
 
        return print_irq_inversion_bug(curr, &root, target_entry,
-                                       this, 1, irqclass);
+                                       this, 0, irqclass);
 }
 
 void print_irqtrace_events(struct task_struct *curr)
@@ -2302,7 +2311,12 @@ void trace_hardirqs_on_caller(unsigned long ip)
                return;
 
        if (unlikely(curr->hardirqs_enabled)) {
-               debug_atomic_inc(&redundant_hardirqs_on);
+               /*
+                * Neither irq nor preemption are disabled here
+                * so this is racy by nature but loosing one hit
+                * in a stat is not a big deal.
+                */
+               __debug_atomic_inc(redundant_hardirqs_on);
                return;
        }
        /* we'll do an OFF -> ON transition: */
@@ -2329,7 +2343,7 @@ void trace_hardirqs_on_caller(unsigned long ip)
 
        curr->hardirq_enable_ip = ip;
        curr->hardirq_enable_event = ++curr->irq_events;
-       debug_atomic_inc(&hardirqs_on_events);
+       debug_atomic_inc(hardirqs_on_events);
 }
 EXPORT_SYMBOL(trace_hardirqs_on_caller);
 
@@ -2361,9 +2375,9 @@ void trace_hardirqs_off_caller(unsigned long ip)
                curr->hardirqs_enabled = 0;
                curr->hardirq_disable_ip = ip;
                curr->hardirq_disable_event = ++curr->irq_events;
-               debug_atomic_inc(&hardirqs_off_events);
+               debug_atomic_inc(hardirqs_off_events);
        } else
-               debug_atomic_inc(&redundant_hardirqs_off);
+               debug_atomic_inc(redundant_hardirqs_off);
 }
 EXPORT_SYMBOL(trace_hardirqs_off_caller);
 
@@ -2387,7 +2401,7 @@ void trace_softirqs_on(unsigned long ip)
                return;
 
        if (curr->softirqs_enabled) {
-               debug_atomic_inc(&redundant_softirqs_on);
+               debug_atomic_inc(redundant_softirqs_on);
                return;
        }
 
@@ -2397,7 +2411,7 @@ void trace_softirqs_on(unsigned long ip)
        curr->softirqs_enabled = 1;
        curr->softirq_enable_ip = ip;
        curr->softirq_enable_event = ++curr->irq_events;
-       debug_atomic_inc(&softirqs_on_events);
+       debug_atomic_inc(softirqs_on_events);
        /*
         * We are going to turn softirqs on, so set the
         * usage bit for all held locks, if hardirqs are
@@ -2427,10 +2441,10 @@ void trace_softirqs_off(unsigned long ip)
                curr->softirqs_enabled = 0;
                curr->softirq_disable_ip = ip;
                curr->softirq_disable_event = ++curr->irq_events;
-               debug_atomic_inc(&softirqs_off_events);
+               debug_atomic_inc(softirqs_off_events);
                DEBUG_LOCKS_WARN_ON(!softirq_count());
        } else
-               debug_atomic_inc(&redundant_softirqs_off);
+               debug_atomic_inc(redundant_softirqs_off);
 }
 
 static void __lockdep_trace_alloc(gfp_t gfp_mask, unsigned long flags)
@@ -2635,7 +2649,7 @@ static int mark_lock(struct task_struct *curr, struct held_lock *this,
                        return 0;
                break;
        case LOCK_USED:
-               debug_atomic_dec(&nr_unused_locks);
+               debug_atomic_dec(nr_unused_locks);
                break;
        default:
                if (!debug_locks_off_graph_unlock())
@@ -2703,13 +2717,15 @@ EXPORT_SYMBOL_GPL(lockdep_init_map);
  */
 static int __lock_acquire(struct lockdep_map *lock, unsigned int subclass,
                          int trylock, int read, int check, int hardirqs_off,
-                         struct lockdep_map *nest_lock, unsigned long ip)
+                         struct lockdep_map *nest_lock, unsigned long ip,
+                         int references)
 {
        struct task_struct *curr = current;
        struct lock_class *class = NULL;
        struct held_lock *hlock;
        unsigned int depth, id;
        int chain_head = 0;
+       int class_idx;
        u64 chain_key;
 
        if (!prove_locking)
@@ -2739,7 +2755,7 @@ static int __lock_acquire(struct lockdep_map *lock, unsigned int subclass,
                if (!class)
                        return 0;
        }
-       debug_atomic_inc((atomic_t *)&class->ops);
+       atomic_inc((atomic_t *)&class->ops);
        if (very_verbose(class)) {
                printk("\nacquire class [%p] %s", class->key, class->name);
                if (class->name_version > 1)
@@ -2757,10 +2773,24 @@ static int __lock_acquire(struct lockdep_map *lock, unsigned int subclass,
        if (DEBUG_LOCKS_WARN_ON(depth >= MAX_LOCK_DEPTH))
                return 0;
 
+       class_idx = class - lock_classes + 1;
+
+       if (depth) {
+               hlock = curr->held_locks + depth - 1;
+               if (hlock->class_idx == class_idx && nest_lock) {
+                       if (hlock->references)
+                               hlock->references++;
+                       else
+                               hlock->references = 2;
+
+                       return 1;
+               }
+       }
+
        hlock = curr->held_locks + depth;
        if (DEBUG_LOCKS_WARN_ON(!class))
                return 0;
-       hlock->class_idx = class - lock_classes + 1;
+       hlock->class_idx = class_idx;
        hlock->acquire_ip = ip;
        hlock->instance = lock;
        hlock->nest_lock = nest_lock;
@@ -2768,9 +2798,10 @@ static int __lock_acquire(struct lockdep_map *lock, unsigned int subclass,
        hlock->read = read;
        hlock->check = check;
        hlock->hardirqs_off = !!hardirqs_off;
+       hlock->references = references;
 #ifdef CONFIG_LOCK_STAT
        hlock->waittime_stamp = 0;
-       hlock->holdtime_stamp = sched_clock();
+       hlock->holdtime_stamp = lockstat_clock();
 #endif
 
        if (check == 2 && !mark_irqflags(curr, hlock))
@@ -2876,6 +2907,30 @@ static int check_unlock(struct task_struct *curr, struct lockdep_map *lock,
        return 1;
 }
 
+static int match_held_lock(struct held_lock *hlock, struct lockdep_map *lock)
+{
+       if (hlock->instance == lock)
+               return 1;
+
+       if (hlock->references) {
+               struct lock_class *class = lock->class_cache;
+
+               if (!class)
+                       class = look_up_lock_class(lock, 0);
+
+               if (DEBUG_LOCKS_WARN_ON(!class))
+                       return 0;
+
+               if (DEBUG_LOCKS_WARN_ON(!hlock->nest_lock))
+                       return 0;
+
+               if (hlock->class_idx == class - lock_classes + 1)
+                       return 1;
+       }
+
+       return 0;
+}
+
 static int
 __lock_set_class(struct lockdep_map *lock, const char *name,
                 struct lock_class_key *key, unsigned int subclass,
@@ -2899,7 +2954,7 @@ __lock_set_class(struct lockdep_map *lock, const char *name,
                 */
                if (prev_hlock && prev_hlock->irq_context != hlock->irq_context)
                        break;
-               if (hlock->instance == lock)
+               if (match_held_lock(hlock, lock))
                        goto found_it;
                prev_hlock = hlock;
        }
@@ -2918,7 +2973,8 @@ found_it:
                if (!__lock_acquire(hlock->instance,
                        hlock_class(hlock)->subclass, hlock->trylock,
                                hlock->read, hlock->check, hlock->hardirqs_off,
-                               hlock->nest_lock, hlock->acquire_ip))
+                               hlock->nest_lock, hlock->acquire_ip,
+                               hlock->references))
                        return 0;
        }
 
@@ -2957,20 +3013,34 @@ lock_release_non_nested(struct task_struct *curr,
                 */
                if (prev_hlock && prev_hlock->irq_context != hlock->irq_context)
                        break;
-               if (hlock->instance == lock)
+               if (match_held_lock(hlock, lock))
                        goto found_it;
                prev_hlock = hlock;
        }
        return print_unlock_inbalance_bug(curr, lock, ip);
 
 found_it:
-       lock_release_holdtime(hlock);
+       if (hlock->instance == lock)
+               lock_release_holdtime(hlock);
+
+       if (hlock->references) {
+               hlock->references--;
+               if (hlock->references) {
+                       /*
+                        * We had, and after removing one, still have
+                        * references, the current lock stack is still
+                        * valid. We're done!
+                        */
+                       return 1;
+               }
+       }
 
        /*
         * We have the right lock to unlock, 'hlock' points to it.
         * Now we remove it from the stack, and add back the other
         * entries (if any), recalculating the hash along the way:
         */
+
        curr->lockdep_depth = i;
        curr->curr_chain_key = hlock->prev_chain_key;
 
@@ -2979,7 +3049,8 @@ found_it:
                if (!__lock_acquire(hlock->instance,
                        hlock_class(hlock)->subclass, hlock->trylock,
                                hlock->read, hlock->check, hlock->hardirqs_off,
-                               hlock->nest_lock, hlock->acquire_ip))
+                               hlock->nest_lock, hlock->acquire_ip,
+                               hlock->references))
                        return 0;
        }
 
@@ -3009,7 +3080,7 @@ static int lock_release_nested(struct task_struct *curr,
        /*
         * Is the unlock non-nested:
         */
-       if (hlock->instance != lock)
+       if (hlock->instance != lock || hlock->references)
                return lock_release_non_nested(curr, lock, ip);
        curr->lockdep_depth--;
 
@@ -3054,6 +3125,21 @@ __lock_release(struct lockdep_map *lock, int nested, unsigned long ip)
        check_chain_key(curr);
 }
 
+static int __lock_is_held(struct lockdep_map *lock)
+{
+       struct task_struct *curr = current;
+       int i;
+
+       for (i = 0; i < curr->lockdep_depth; i++) {
+               struct held_lock *hlock = curr->held_locks + i;
+
+               if (match_held_lock(hlock, lock))
+                       return 1;
+       }
+
+       return 0;
+}
+
 /*
  * Check whether we follow the irq-flags state precisely:
  */
@@ -3120,8 +3206,6 @@ void lock_acquire(struct lockdep_map *lock, unsigned int subclass,
 {
        unsigned long flags;
 
-       trace_lock_acquire(lock, subclass, trylock, read, check, nest_lock, ip);
-
        if (unlikely(current->lockdep_recursion))
                return;
 
@@ -3129,8 +3213,9 @@ void lock_acquire(struct lockdep_map *lock, unsigned int subclass,
        check_flags(flags);
 
        current->lockdep_recursion = 1;
+       trace_lock_acquire(lock, subclass, trylock, read, check, nest_lock, ip);
        __lock_acquire(lock, subclass, trylock, read, check,
-                      irqs_disabled_flags(flags), nest_lock, ip);
+                      irqs_disabled_flags(flags), nest_lock, ip, 0);
        current->lockdep_recursion = 0;
        raw_local_irq_restore(flags);
 }
@@ -3141,20 +3226,39 @@ void lock_release(struct lockdep_map *lock, int nested,
 {
        unsigned long flags;
 
-       trace_lock_release(lock, nested, ip);
-
        if (unlikely(current->lockdep_recursion))
                return;
 
        raw_local_irq_save(flags);
        check_flags(flags);
        current->lockdep_recursion = 1;
+       trace_lock_release(lock, ip);
        __lock_release(lock, nested, ip);
        current->lockdep_recursion = 0;
        raw_local_irq_restore(flags);
 }
 EXPORT_SYMBOL_GPL(lock_release);
 
+int lock_is_held(struct lockdep_map *lock)
+{
+       unsigned long flags;
+       int ret = 0;
+
+       if (unlikely(current->lockdep_recursion))
+               return ret;
+
+       raw_local_irq_save(flags);
+       check_flags(flags);
+
+       current->lockdep_recursion = 1;
+       ret = __lock_is_held(lock);
+       current->lockdep_recursion = 0;
+       raw_local_irq_restore(flags);
+
+       return ret;
+}
+EXPORT_SYMBOL_GPL(lock_is_held);
+
 void lockdep_set_current_reclaim_state(gfp_t gfp_mask)
 {
        current->lockdep_reclaim_gfp = gfp_mask;
@@ -3214,7 +3318,7 @@ __lock_contended(struct lockdep_map *lock, unsigned long ip)
                 */
                if (prev_hlock && prev_hlock->irq_context != hlock->irq_context)
                        break;
-               if (hlock->instance == lock)
+               if (match_held_lock(hlock, lock))
                        goto found_it;
                prev_hlock = hlock;
        }
@@ -3222,7 +3326,10 @@ __lock_contended(struct lockdep_map *lock, unsigned long ip)
        return;
 
 found_it:
-       hlock->waittime_stamp = sched_clock();
+       if (hlock->instance != lock)
+               return;
+
+       hlock->waittime_stamp = lockstat_clock();
 
        contention_point = lock_point(hlock_class(hlock)->contention_point, ip);
        contending_point = lock_point(hlock_class(hlock)->contending_point,
@@ -3245,8 +3352,7 @@ __lock_acquired(struct lockdep_map *lock, unsigned long ip)
        struct held_lock *hlock, *prev_hlock;
        struct lock_class_stats *stats;
        unsigned int depth;
-       u64 now;
-       s64 waittime = 0;
+       u64 now, waittime = 0;
        int i, cpu;
 
        depth = curr->lockdep_depth;
@@ -3261,7 +3367,7 @@ __lock_acquired(struct lockdep_map *lock, unsigned long ip)
                 */
                if (prev_hlock && prev_hlock->irq_context != hlock->irq_context)
                        break;
-               if (hlock->instance == lock)
+               if (match_held_lock(hlock, lock))
                        goto found_it;
                prev_hlock = hlock;
        }
@@ -3269,14 +3375,17 @@ __lock_acquired(struct lockdep_map *lock, unsigned long ip)
        return;
 
 found_it:
+       if (hlock->instance != lock)
+               return;
+
        cpu = smp_processor_id();
        if (hlock->waittime_stamp) {
-               now = sched_clock();
+               now = lockstat_clock();
                waittime = now - hlock->waittime_stamp;
                hlock->holdtime_stamp = now;
        }
 
-       trace_lock_acquired(lock, ip, waittime);
+       trace_lock_acquired(lock, ip);
 
        stats = get_lock_stats(hlock_class(hlock));
        if (waittime) {
@@ -3297,8 +3406,6 @@ void lock_contended(struct lockdep_map *lock, unsigned long ip)
 {
        unsigned long flags;
 
-       trace_lock_contended(lock, ip);
-
        if (unlikely(!lock_stat))
                return;
 
@@ -3308,6 +3415,7 @@ void lock_contended(struct lockdep_map *lock, unsigned long ip)
        raw_local_irq_save(flags);
        check_flags(flags);
        current->lockdep_recursion = 1;
+       trace_lock_contended(lock, ip);
        __lock_contended(lock, ip);
        current->lockdep_recursion = 0;
        raw_local_irq_restore(flags);
@@ -3499,10 +3607,11 @@ void __init lockdep_info(void)
                sizeof(struct list_head) * CLASSHASH_SIZE +
                sizeof(struct lock_list) * MAX_LOCKDEP_ENTRIES +
                sizeof(struct lock_chain) * MAX_LOCKDEP_CHAINS +
-               sizeof(struct list_head) * CHAINHASH_SIZE) / 1024
+               sizeof(struct list_head) * CHAINHASH_SIZE
 #ifdef CONFIG_PROVE_LOCKING
-               + sizeof(struct circular_queue) + sizeof(bfs_accessed)
+               + sizeof(struct circular_queue)
 #endif
+               ) / 1024
                );
 
        printk(" per task-struct memory footprint: %lu bytes\n",
@@ -3692,3 +3801,25 @@ void lockdep_sys_exit(void)
                lockdep_print_held_locks(curr);
        }
 }
+
+void lockdep_rcu_dereference(const char *file, const int line)
+{
+       struct task_struct *curr = current;
+
+#ifndef CONFIG_PROVE_RCU_REPEATEDLY
+       if (!debug_locks_off())
+               return;
+#endif /* #ifdef CONFIG_PROVE_RCU_REPEATEDLY */
+       /* Note: the following can be executed concurrently, so be careful. */
+       printk("\n===================================================\n");
+       printk(  "[ INFO: suspicious rcu_dereference_check() usage. ]\n");
+       printk(  "---------------------------------------------------\n");
+       printk("%s:%d invoked rcu_dereference_check() without protection!\n",
+                       file, line);
+       printk("\nother info that might help us debug this:\n\n");
+       printk("\nrcu_scheduler_active = %d, debug_locks = %d\n", rcu_scheduler_active, debug_locks);
+       lockdep_print_held_locks(curr);
+       printk("\nstack backtrace:\n");
+       dump_stack();
+}
+EXPORT_SYMBOL_GPL(lockdep_rcu_dereference);