4 * The ldisc semaphore is semantically a rw_semaphore but which enforces
5 * an alternate policy, namely:
6 * 1) Supports lock wait timeouts
7 * 2) Write waiter has priority
8 * 3) Downgrading is not supported
10 * Implementation notes:
11 * 1) Upper half of semaphore count is a wait count (differs from rwsem
12 * in that rwsem normalizes the upper half to the wait bias)
13 * 2) Lacks overflow checking
15 * The generic counting was copied and modified from include/asm-generic/rwsem.h
18 * The scheduling policy was copied and modified from lib/rwsem.c
21 * This implementation incorporates the write lock stealing work of
26 * This file may be redistributed under the terms of the GNU General Public
30 #include <linux/list.h>
31 #include <linux/spinlock.h>
32 #include <linux/atomic.h>
33 #include <linux/tty.h>
34 #include <linux/sched.h>
37 #ifdef CONFIG_DEBUG_LOCK_ALLOC
38 # define __acq(l, s, t, r, c, n, i) \
39 lock_acquire(&(l)->dep_map, s, t, r, c, n, i)
40 # define __rel(l, n, i) \
41 lock_release(&(l)->dep_map, n, i)
42 #define lockdep_acquire(l, s, t, i) __acq(l, s, t, 0, 1, NULL, i)
43 #define lockdep_acquire_nest(l, s, t, n, i) __acq(l, s, t, 0, 1, n, i)
44 #define lockdep_acquire_read(l, s, t, i) __acq(l, s, t, 1, 1, NULL, i)
45 #define lockdep_release(l, n, i) __rel(l, n, i)
47 # define lockdep_acquire(l, s, t, i) do { } while (0)
48 # define lockdep_acquire_nest(l, s, t, n, i) do { } while (0)
49 # define lockdep_acquire_read(l, s, t, i) do { } while (0)
50 # define lockdep_release(l, n, i) do { } while (0)
53 #ifdef CONFIG_LOCK_STAT
54 # define lock_stat(_lock, stat) lock_##stat(&(_lock)->dep_map, _RET_IP_)
56 # define lock_stat(_lock, stat) do { } while (0)
60 #if BITS_PER_LONG == 64
61 # define LDSEM_ACTIVE_MASK 0xffffffffL
63 # define LDSEM_ACTIVE_MASK 0x0000ffffL
66 #define LDSEM_UNLOCKED 0L
67 #define LDSEM_ACTIVE_BIAS 1L
68 #define LDSEM_WAIT_BIAS (-LDSEM_ACTIVE_MASK-1)
69 #define LDSEM_READ_BIAS LDSEM_ACTIVE_BIAS
70 #define LDSEM_WRITE_BIAS (LDSEM_WAIT_BIAS + LDSEM_ACTIVE_BIAS)
73 struct list_head list;
74 struct task_struct *task;
77 static inline long ldsem_atomic_update(long delta, struct ld_semaphore *sem)
79 return atomic_long_add_return(delta, (atomic_long_t *)&sem->count);
83 * ldsem_cmpxchg() updates @*old with the last-known sem->count value.
84 * Returns 1 if count was successfully changed; @*old will have @new value.
85 * Returns 0 if count was not changed; @*old will have most recent sem->count
87 static inline int ldsem_cmpxchg(long *old, long new, struct ld_semaphore *sem)
89 long tmp = atomic_long_cmpxchg(&sem->count, *old, new);
100 * Initialize an ldsem:
102 void __init_ldsem(struct ld_semaphore *sem, const char *name,
103 struct lock_class_key *key)
105 #ifdef CONFIG_DEBUG_LOCK_ALLOC
107 * Make sure we are not reinitializing a held semaphore:
109 debug_check_no_locks_freed((void *)sem, sizeof(*sem));
110 lockdep_init_map(&sem->dep_map, name, key, 0);
112 sem->count = LDSEM_UNLOCKED;
113 sem->wait_readers = 0;
114 raw_spin_lock_init(&sem->wait_lock);
115 INIT_LIST_HEAD(&sem->read_wait);
116 INIT_LIST_HEAD(&sem->write_wait);
119 static void __ldsem_wake_readers(struct ld_semaphore *sem)
121 struct ldsem_waiter *waiter, *next;
122 struct task_struct *tsk;
125 /* Try to grant read locks to all readers on the read wait list.
126 * Note the 'active part' of the count is incremented by
127 * the number of readers before waking any processes up.
129 adjust = sem->wait_readers * (LDSEM_ACTIVE_BIAS - LDSEM_WAIT_BIAS);
130 count = ldsem_atomic_update(adjust, sem);
134 if (ldsem_cmpxchg(&count, count - adjust, sem))
138 list_for_each_entry_safe(waiter, next, &sem->read_wait, list) {
142 wake_up_process(tsk);
143 put_task_struct(tsk);
145 INIT_LIST_HEAD(&sem->read_wait);
146 sem->wait_readers = 0;
149 static inline int writer_trylock(struct ld_semaphore *sem)
151 /* only wake this writer if the active part of the count can be
152 * transitioned from 0 -> 1
154 long count = ldsem_atomic_update(LDSEM_ACTIVE_BIAS, sem);
156 if ((count & LDSEM_ACTIVE_MASK) == LDSEM_ACTIVE_BIAS)
158 if (ldsem_cmpxchg(&count, count - LDSEM_ACTIVE_BIAS, sem))
163 static void __ldsem_wake_writer(struct ld_semaphore *sem)
165 struct ldsem_waiter *waiter;
167 waiter = list_entry(sem->write_wait.next, struct ldsem_waiter, list);
168 wake_up_process(waiter->task);
172 * handle the lock release when processes blocked on it that can now run
173 * - if we come here from up_xxxx(), then:
174 * - the 'active part' of count (&0x0000ffff) reached 0 (but may have changed)
175 * - the 'waiting part' of count (&0xffff0000) is -ve (and will still be so)
176 * - the spinlock must be held by the caller
177 * - woken process blocks are discarded from the list after having task zeroed
179 static void __ldsem_wake(struct ld_semaphore *sem)
181 if (!list_empty(&sem->write_wait))
182 __ldsem_wake_writer(sem);
183 else if (!list_empty(&sem->read_wait))
184 __ldsem_wake_readers(sem);
187 static void ldsem_wake(struct ld_semaphore *sem)
191 raw_spin_lock_irqsave(&sem->wait_lock, flags);
193 raw_spin_unlock_irqrestore(&sem->wait_lock, flags);
197 * wait for the read lock to be granted
199 static struct ld_semaphore __sched *
200 down_read_failed(struct ld_semaphore *sem, long count, long timeout)
202 struct ldsem_waiter waiter;
203 struct task_struct *tsk = current;
204 long adjust = -LDSEM_ACTIVE_BIAS + LDSEM_WAIT_BIAS;
206 /* set up my own style of waitqueue */
207 raw_spin_lock_irq(&sem->wait_lock);
209 /* Try to reverse the lock attempt but if the count has changed
210 * so that reversing fails, check if there are are no waiters,
211 * and early-out if not */
213 if (ldsem_cmpxchg(&count, count + adjust, sem))
216 raw_spin_unlock_irq(&sem->wait_lock);
221 list_add_tail(&waiter.list, &sem->read_wait);
225 get_task_struct(tsk);
227 /* if there are no active locks, wake the new lock owner(s) */
228 if ((count & LDSEM_ACTIVE_MASK) == 0)
231 raw_spin_unlock_irq(&sem->wait_lock);
233 /* wait to be given the lock */
235 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
241 timeout = schedule_timeout(timeout);
244 __set_task_state(tsk, TASK_RUNNING);
247 /* lock timed out but check if this task was just
248 * granted lock ownership - if so, pretend there
249 * was no timeout; otherwise, cleanup lock wait */
250 raw_spin_lock_irq(&sem->wait_lock);
252 ldsem_atomic_update(-LDSEM_WAIT_BIAS, sem);
253 list_del(&waiter.list);
254 raw_spin_unlock_irq(&sem->wait_lock);
255 put_task_struct(waiter.task);
258 raw_spin_unlock_irq(&sem->wait_lock);
265 * wait for the write lock to be granted
267 static struct ld_semaphore __sched *
268 down_write_failed(struct ld_semaphore *sem, long count, long timeout)
270 struct ldsem_waiter waiter;
271 struct task_struct *tsk = current;
272 long adjust = -LDSEM_ACTIVE_BIAS;
275 /* set up my own style of waitqueue */
276 raw_spin_lock_irq(&sem->wait_lock);
278 /* Try to reverse the lock attempt but if the count has changed
279 * so that reversing fails, check if the lock is now owned,
280 * and early-out if so */
282 if (ldsem_cmpxchg(&count, count + adjust, sem))
284 if ((count & LDSEM_ACTIVE_MASK) == LDSEM_ACTIVE_BIAS) {
285 raw_spin_unlock_irq(&sem->wait_lock);
290 list_add_tail(&waiter.list, &sem->write_wait);
294 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
298 raw_spin_unlock_irq(&sem->wait_lock);
299 timeout = schedule_timeout(timeout);
300 raw_spin_lock_irq(&sem->wait_lock);
301 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
302 if ((locked = writer_trylock(sem)))
307 ldsem_atomic_update(-LDSEM_WAIT_BIAS, sem);
308 list_del(&waiter.list);
309 raw_spin_unlock_irq(&sem->wait_lock);
311 __set_task_state(tsk, TASK_RUNNING);
313 /* lock wait may have timed out */
321 static inline int __ldsem_down_read_nested(struct ld_semaphore *sem,
322 int subclass, long timeout)
326 lockdep_acquire_read(sem, subclass, 0, _RET_IP_);
328 count = ldsem_atomic_update(LDSEM_READ_BIAS, sem);
330 lock_stat(sem, contended);
331 if (!down_read_failed(sem, count, timeout)) {
332 lockdep_release(sem, 1, _RET_IP_);
336 lock_stat(sem, acquired);
340 static inline int __ldsem_down_write_nested(struct ld_semaphore *sem,
341 int subclass, long timeout)
345 lockdep_acquire(sem, subclass, 0, _RET_IP_);
347 count = ldsem_atomic_update(LDSEM_WRITE_BIAS, sem);
348 if ((count & LDSEM_ACTIVE_MASK) != LDSEM_ACTIVE_BIAS) {
349 lock_stat(sem, contended);
350 if (!down_write_failed(sem, count, timeout)) {
351 lockdep_release(sem, 1, _RET_IP_);
355 lock_stat(sem, acquired);
361 * lock for reading -- returns 1 if successful, 0 if timed out
363 int __sched ldsem_down_read(struct ld_semaphore *sem, long timeout)
366 return __ldsem_down_read_nested(sem, 0, timeout);
370 * trylock for reading -- returns 1 if successful, 0 if contention
372 int ldsem_down_read_trylock(struct ld_semaphore *sem)
374 long count = sem->count;
377 if (ldsem_cmpxchg(&count, count + LDSEM_READ_BIAS, sem)) {
378 lockdep_acquire_read(sem, 0, 1, _RET_IP_);
379 lock_stat(sem, acquired);
387 * lock for writing -- returns 1 if successful, 0 if timed out
389 int __sched ldsem_down_write(struct ld_semaphore *sem, long timeout)
392 return __ldsem_down_write_nested(sem, 0, timeout);
396 * trylock for writing -- returns 1 if successful, 0 if contention
398 int ldsem_down_write_trylock(struct ld_semaphore *sem)
400 long count = sem->count;
402 while ((count & LDSEM_ACTIVE_MASK) == 0) {
403 if (ldsem_cmpxchg(&count, count + LDSEM_WRITE_BIAS, sem)) {
404 lockdep_acquire(sem, 0, 1, _RET_IP_);
405 lock_stat(sem, acquired);
413 * release a read lock
415 void ldsem_up_read(struct ld_semaphore *sem)
419 lockdep_release(sem, 1, _RET_IP_);
421 count = ldsem_atomic_update(-LDSEM_READ_BIAS, sem);
422 if (count < 0 && (count & LDSEM_ACTIVE_MASK) == 0)
427 * release a write lock
429 void ldsem_up_write(struct ld_semaphore *sem)
433 lockdep_release(sem, 1, _RET_IP_);
435 count = ldsem_atomic_update(-LDSEM_WRITE_BIAS, sem);
441 #ifdef CONFIG_DEBUG_LOCK_ALLOC
443 int ldsem_down_read_nested(struct ld_semaphore *sem, int subclass, long timeout)
446 return __ldsem_down_read_nested(sem, subclass, timeout);
449 int ldsem_down_write_nested(struct ld_semaphore *sem, int subclass,
453 return __ldsem_down_write_nested(sem, subclass, timeout);