diff options
| author | Ming Lei <tom.leiming@gmail.com> | 2009-07-16 09:44:29 -0400 |
|---|---|---|
| committer | Peter Zijlstra <a.p.zijlstra@chello.nl> | 2009-07-24 04:49:46 -0400 |
| commit | d588e46155e9c51217b9840db1e94a0f594c1af2 (patch) | |
| tree | 0ede7d7d8338f8db15d8c690c46a0173196d7bac /kernel | |
| parent | c94aa5ca3088018d2a7a9bd3258aefffe29df265 (diff) | |
lockdep: Improve implementation of BFS
1,replace %MAX_CIRCULAR_QUE_SIZE with &(MAX_CIRCULAR_QUE_SIZE-1)
since we define MAX_CIRCULAR_QUE_SIZE as power of 2;
2,use bitmap to mark if a lock is accessed in BFS in order to
clear it quickly, because we may search a graph many times.
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1246201486-7308-3-git-send-email-tom.leiming@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Diffstat (limited to 'kernel')
| -rw-r--r-- | kernel/lockdep.c | 23 | ||||
| -rw-r--r-- | kernel/lockdep_internals.h | 35 |
2 files changed, 39 insertions, 19 deletions
diff --git a/kernel/lockdep.c b/kernel/lockdep.c index 93dc70d18cdf..5dcca26a8263 100644 --- a/kernel/lockdep.c +++ b/kernel/lockdep.c | |||
| @@ -42,7 +42,7 @@ | |||
| 42 | #include <linux/hash.h> | 42 | #include <linux/hash.h> |
| 43 | #include <linux/ftrace.h> | 43 | #include <linux/ftrace.h> |
| 44 | #include <linux/stringify.h> | 44 | #include <linux/stringify.h> |
| 45 | 45 | #include <linux/bitops.h> | |
| 46 | #include <asm/sections.h> | 46 | #include <asm/sections.h> |
| 47 | 47 | ||
| 48 | #include "lockdep_internals.h" | 48 | #include "lockdep_internals.h" |
| @@ -118,7 +118,7 @@ static inline int debug_locks_off_graph_unlock(void) | |||
| 118 | static int lockdep_initialized; | 118 | static int lockdep_initialized; |
| 119 | 119 | ||
| 120 | unsigned long nr_list_entries; | 120 | unsigned long nr_list_entries; |
| 121 | static struct lock_list list_entries[MAX_LOCKDEP_ENTRIES]; | 121 | struct lock_list list_entries[MAX_LOCKDEP_ENTRIES]; |
| 122 | 122 | ||
| 123 | /* | 123 | /* |
| 124 | * All data structures here are protected by the global debug_lock. | 124 | * All data structures here are protected by the global debug_lock. |
| @@ -897,30 +897,38 @@ static int add_lock_to_list(struct lock_class *class, struct lock_class *this, | |||
| 897 | return 1; | 897 | return 1; |
| 898 | } | 898 | } |
| 899 | 899 | ||
| 900 | unsigned long bfs_accessed[BITS_TO_LONGS(MAX_LOCKDEP_ENTRIES)]; | ||
| 900 | static struct circular_queue lock_cq; | 901 | static struct circular_queue lock_cq; |
| 902 | |||
| 901 | static int __search_shortest_path(struct lock_list *source_entry, | 903 | static int __search_shortest_path(struct lock_list *source_entry, |
| 902 | struct lock_class *target, | 904 | struct lock_class *target, |
| 903 | struct lock_list **target_entry, | 905 | struct lock_list **target_entry, |
| 904 | int forward) | 906 | int forward) |
| 905 | { | 907 | { |
| 906 | struct lock_list *entry; | 908 | struct lock_list *entry; |
| 909 | struct list_head *head; | ||
| 907 | struct circular_queue *cq = &lock_cq; | 910 | struct circular_queue *cq = &lock_cq; |
| 908 | int ret = 1; | 911 | int ret = 1; |
| 909 | 912 | ||
| 910 | __cq_init(cq); | ||
| 911 | |||
| 912 | mark_lock_accessed(source_entry, NULL); | ||
| 913 | if (source_entry->class == target) { | 913 | if (source_entry->class == target) { |
| 914 | *target_entry = source_entry; | 914 | *target_entry = source_entry; |
| 915 | ret = 0; | 915 | ret = 0; |
| 916 | goto exit; | 916 | goto exit; |
| 917 | } | 917 | } |
| 918 | 918 | ||
| 919 | if (forward) | ||
| 920 | head = &source_entry->class->locks_after; | ||
| 921 | else | ||
| 922 | head = &source_entry->class->locks_before; | ||
| 923 | |||
| 924 | if (list_empty(head)) | ||
| 925 | goto exit; | ||
| 926 | |||
| 927 | __cq_init(cq); | ||
| 919 | __cq_enqueue(cq, (unsigned long)source_entry); | 928 | __cq_enqueue(cq, (unsigned long)source_entry); |
| 920 | 929 | ||
| 921 | while (!__cq_empty(cq)) { | 930 | while (!__cq_empty(cq)) { |
| 922 | struct lock_list *lock; | 931 | struct lock_list *lock; |
| 923 | struct list_head *head; | ||
| 924 | 932 | ||
| 925 | __cq_dequeue(cq, (unsigned long *)&lock); | 933 | __cq_dequeue(cq, (unsigned long *)&lock); |
| 926 | 934 | ||
| @@ -1040,6 +1048,7 @@ static noinline int print_circular_bug(void) | |||
| 1040 | return 0; | 1048 | return 0; |
| 1041 | 1049 | ||
| 1042 | this.class = hlock_class(check_source); | 1050 | this.class = hlock_class(check_source); |
| 1051 | this.parent = NULL; | ||
| 1043 | if (!save_trace(&this.trace)) | 1052 | if (!save_trace(&this.trace)) |
| 1044 | return 0; | 1053 | return 0; |
| 1045 | 1054 | ||
| @@ -1580,10 +1589,10 @@ check_prev_add(struct task_struct *curr, struct held_lock *prev, | |||
| 1580 | */ | 1589 | */ |
| 1581 | check_source = next; | 1590 | check_source = next; |
| 1582 | check_target = prev; | 1591 | check_target = prev; |
| 1592 | |||
| 1583 | if (check_noncircular(hlock_class(next), 0) == 2) | 1593 | if (check_noncircular(hlock_class(next), 0) == 2) |
| 1584 | return print_circular_bug(); | 1594 | return print_circular_bug(); |
| 1585 | 1595 | ||
| 1586 | |||
| 1587 | if (!check_prev_add_irq(curr, prev, next)) | 1596 | if (!check_prev_add_irq(curr, prev, next)) |
| 1588 | return 0; | 1597 | return 0; |
| 1589 | 1598 | ||
diff --git a/kernel/lockdep_internals.h b/kernel/lockdep_internals.h index 6f48d37d5be2..c2f6594966f3 100644 --- a/kernel/lockdep_internals.h +++ b/kernel/lockdep_internals.h | |||
| @@ -137,23 +137,28 @@ extern atomic_t nr_find_usage_backwards_recursions; | |||
| 137 | # define debug_atomic_read(ptr) 0 | 137 | # define debug_atomic_read(ptr) 0 |
| 138 | #endif | 138 | #endif |
| 139 | 139 | ||
| 140 | |||
| 141 | extern unsigned long nr_list_entries; | ||
| 142 | extern struct lock_list list_entries[MAX_LOCKDEP_ENTRIES]; | ||
| 143 | extern unsigned long bfs_accessed[]; | ||
| 144 | |||
| 145 | /*For good efficiency of modular, we use power of 2*/ | ||
| 146 | #define MAX_CIRCULAR_QUE_SIZE 4096UL | ||
| 147 | |||
| 140 | /* The circular_queue and helpers is used to implement the | 148 | /* The circular_queue and helpers is used to implement the |
| 141 | * breadth-first search(BFS)algorithem, by which we can build | 149 | * breadth-first search(BFS)algorithem, by which we can build |
| 142 | * the shortest path from the next lock to be acquired to the | 150 | * the shortest path from the next lock to be acquired to the |
| 143 | * previous held lock if there is a circular between them. | 151 | * previous held lock if there is a circular between them. |
| 144 | * */ | 152 | * */ |
| 145 | #define MAX_CIRCULAR_QUE_SIZE 4096UL | ||
| 146 | struct circular_queue{ | 153 | struct circular_queue{ |
| 147 | unsigned long element[MAX_CIRCULAR_QUE_SIZE]; | 154 | unsigned long element[MAX_CIRCULAR_QUE_SIZE]; |
| 148 | unsigned int front, rear; | 155 | unsigned int front, rear; |
| 149 | }; | 156 | }; |
| 150 | 157 | ||
| 151 | #define LOCK_ACCESSED 1UL | ||
| 152 | #define LOCK_ACCESSED_MASK (~LOCK_ACCESSED) | ||
| 153 | |||
| 154 | static inline void __cq_init(struct circular_queue *cq) | 158 | static inline void __cq_init(struct circular_queue *cq) |
| 155 | { | 159 | { |
| 156 | cq->front = cq->rear = 0; | 160 | cq->front = cq->rear = 0; |
| 161 | bitmap_zero(bfs_accessed, MAX_LOCKDEP_ENTRIES); | ||
| 157 | } | 162 | } |
| 158 | 163 | ||
| 159 | static inline int __cq_empty(struct circular_queue *cq) | 164 | static inline int __cq_empty(struct circular_queue *cq) |
| @@ -163,7 +168,7 @@ static inline int __cq_empty(struct circular_queue *cq) | |||
| 163 | 168 | ||
| 164 | static inline int __cq_full(struct circular_queue *cq) | 169 | static inline int __cq_full(struct circular_queue *cq) |
| 165 | { | 170 | { |
| 166 | return ((cq->rear + 1)%MAX_CIRCULAR_QUE_SIZE) == cq->front; | 171 | return ((cq->rear + 1)&(MAX_CIRCULAR_QUE_SIZE-1)) == cq->front; |
| 167 | } | 172 | } |
| 168 | 173 | ||
| 169 | static inline int __cq_enqueue(struct circular_queue *cq, unsigned long elem) | 174 | static inline int __cq_enqueue(struct circular_queue *cq, unsigned long elem) |
| @@ -172,7 +177,7 @@ static inline int __cq_enqueue(struct circular_queue *cq, unsigned long elem) | |||
| 172 | return -1; | 177 | return -1; |
| 173 | 178 | ||
| 174 | cq->element[cq->rear] = elem; | 179 | cq->element[cq->rear] = elem; |
| 175 | cq->rear = (cq->rear + 1)%MAX_CIRCULAR_QUE_SIZE; | 180 | cq->rear = (cq->rear + 1)&(MAX_CIRCULAR_QUE_SIZE-1); |
| 176 | return 0; | 181 | return 0; |
| 177 | } | 182 | } |
| 178 | 183 | ||
| @@ -182,30 +187,36 @@ static inline int __cq_dequeue(struct circular_queue *cq, unsigned long *elem) | |||
| 182 | return -1; | 187 | return -1; |
| 183 | 188 | ||
| 184 | *elem = cq->element[cq->front]; | 189 | *elem = cq->element[cq->front]; |
| 185 | cq->front = (cq->front + 1)%MAX_CIRCULAR_QUE_SIZE; | 190 | cq->front = (cq->front + 1)&(MAX_CIRCULAR_QUE_SIZE-1); |
| 186 | return 0; | 191 | return 0; |
| 187 | } | 192 | } |
| 188 | 193 | ||
| 189 | static inline int __cq_get_elem_count(struct circular_queue *cq) | 194 | static inline int __cq_get_elem_count(struct circular_queue *cq) |
| 190 | { | 195 | { |
| 191 | return (cq->rear - cq->front)%MAX_CIRCULAR_QUE_SIZE; | 196 | return (cq->rear - cq->front)&(MAX_CIRCULAR_QUE_SIZE-1); |
| 192 | } | 197 | } |
| 193 | 198 | ||
| 194 | static inline void mark_lock_accessed(struct lock_list *lock, | 199 | static inline void mark_lock_accessed(struct lock_list *lock, |
| 195 | struct lock_list *parent) | 200 | struct lock_list *parent) |
| 196 | { | 201 | { |
| 197 | lock->parent = (void *) parent + LOCK_ACCESSED; | 202 | unsigned long nr; |
| 203 | nr = lock - list_entries; | ||
| 204 | WARN_ON(nr >= nr_list_entries); | ||
| 205 | lock->parent = parent; | ||
| 206 | set_bit(nr, bfs_accessed); | ||
| 198 | } | 207 | } |
| 199 | 208 | ||
| 200 | static inline unsigned long lock_accessed(struct lock_list *lock) | 209 | static inline unsigned long lock_accessed(struct lock_list *lock) |
| 201 | { | 210 | { |
| 202 | return (unsigned long)lock->parent & LOCK_ACCESSED; | 211 | unsigned long nr; |
| 212 | nr = lock - list_entries; | ||
| 213 | WARN_ON(nr >= nr_list_entries); | ||
| 214 | return test_bit(nr, bfs_accessed); | ||
| 203 | } | 215 | } |
| 204 | 216 | ||
| 205 | static inline struct lock_list *get_lock_parent(struct lock_list *child) | 217 | static inline struct lock_list *get_lock_parent(struct lock_list *child) |
| 206 | { | 218 | { |
| 207 | return (struct lock_list *) | 219 | return child->parent; |
| 208 | ((unsigned long)child->parent & LOCK_ACCESSED_MASK); | ||
| 209 | } | 220 | } |
| 210 | 221 | ||
| 211 | static inline unsigned long get_lock_depth(struct lock_list *child) | 222 | static inline unsigned long get_lock_depth(struct lock_list *child) |
