1 /* rwsem.c: R/W semaphores: contention handling functions
3 * Written by David Howells (dhowells@redhat.com).
4 * Derived from arch/i386/kernel/semaphore.c
6 #include <linux/rwsem.h>
7 #include <linux/sched.h>
8 #include <linux/init.h>
9 #include <linux/module.h>
12 * Initialize an rwsem:
14 void __init_rwsem(struct rw_semaphore *sem, const char *name,
15 struct lock_class_key *key)
17 #ifdef CONFIG_DEBUG_LOCK_ALLOC
19 * Make sure we are not reinitializing a held semaphore:
21 debug_check_no_locks_freed((void *)sem, sizeof(*sem));
22 lockdep_init_map(&sem->dep_map, name, key, 0);
24 sem->count = RWSEM_UNLOCKED_VALUE;
25 spin_lock_init(&sem->wait_lock);
26 INIT_LIST_HEAD(&sem->wait_list);
29 EXPORT_SYMBOL(__init_rwsem);
32 struct list_head list;
33 struct task_struct *task;
35 #define RWSEM_WAITING_FOR_READ 0x00000001
36 #define RWSEM_WAITING_FOR_WRITE 0x00000002
39 /* Wake types for __rwsem_do_wake(). Note that RWSEM_WAKE_NO_ACTIVE and
40 * RWSEM_WAKE_READ_OWNED imply that the spinlock must have been kept held
41 * since the rwsem value was observed.
43 #define RWSEM_WAKE_ANY 0 /* Wake whatever's at head of wait list */
44 #define RWSEM_WAKE_NO_ACTIVE 1 /* rwsem was observed with no active thread */
45 #define RWSEM_WAKE_READ_OWNED 2 /* rwsem was observed to be read owned */
48 * handle the lock release when processes blocked on it that can now run
49 * - if we come here from up_xxxx(), then:
50 * - the 'active part' of count (&0x0000ffff) reached 0 (but may have changed)
51 * - the 'waiting part' of count (&0xffff0000) is -ve (and will still be so)
52 * - there must be someone on the queue
53 * - the spinlock must be held by the caller
54 * - woken process blocks are discarded from the list after having task zeroed
55 * - writers are only woken if downgrading is false
57 static struct rw_semaphore *
58 __rwsem_do_wake(struct rw_semaphore *sem, int wake_type)
60 struct rwsem_waiter *waiter;
61 struct task_struct *tsk;
62 struct list_head *next;
63 signed long oldcount, woken, loop;
65 waiter = list_entry(sem->wait_list.next, struct rwsem_waiter, list);
66 if (!(waiter->flags & RWSEM_WAITING_FOR_WRITE))
69 if (wake_type == RWSEM_WAKE_READ_OWNED)
72 /* There's a writer at the front of the queue - try to grant it the
73 * write lock. However, we only wake this writer if we can transition
74 * the active part of the count from 0 -> 1
77 oldcount = rwsem_atomic_update(RWSEM_ACTIVE_BIAS, sem)
79 if (oldcount & RWSEM_ACTIVE_MASK)
80 /* Someone grabbed the sem already */
83 /* We must be careful not to touch 'waiter' after we set ->task = NULL.
84 * It is an allocated on the waiter's stack and may become invalid at
85 * any time after that point (due to a wakeup from another source).
87 list_del(&waiter->list);
96 /* If we come here from up_xxxx(), another thread might have reached
97 * rwsem_down_failed_common() before we acquired the spinlock and
98 * woken up a waiter, making it now active. We prefer to check for
99 * this first in order to not spend too much time with the spinlock
100 * held if we're not going to be able to wake up readers in the end.
102 * Note that we do not need to update the rwsem count: any writer
103 * trying to acquire rwsem will run rwsem_down_write_failed() due
104 * to the waiting threads and block trying to acquire the spinlock.
106 * We use a dummy atomic update in order to acquire the cache line
107 * exclusively since we expect to succeed and run the final rwsem
108 * count adjustment pretty soon.
110 if (wake_type == RWSEM_WAKE_ANY &&
111 (rwsem_atomic_update(0, sem) & RWSEM_ACTIVE_MASK))
112 /* Someone grabbed the sem already */
115 /* Grant an infinite number of read locks to the readers at the front
116 * of the queue. Note we increment the 'active part' of the count by
117 * the number of readers before waking any processes up.
123 if (waiter->list.next == &sem->wait_list)
126 waiter = list_entry(waiter->list.next,
127 struct rwsem_waiter, list);
129 } while (waiter->flags & RWSEM_WAITING_FOR_READ);
132 woken *= RWSEM_ACTIVE_BIAS - RWSEM_WAITING_BIAS;
134 rwsem_atomic_add(woken, sem);
136 next = sem->wait_list.next;
137 for (; loop > 0; loop--) {
138 waiter = list_entry(next, struct rwsem_waiter, list);
139 next = waiter->list.next;
143 wake_up_process(tsk);
144 put_task_struct(tsk);
147 sem->wait_list.next = next;
148 next->prev = &sem->wait_list;
153 /* undo the change to the active count, but check for a transition
156 if (rwsem_atomic_update(-RWSEM_ACTIVE_BIAS, sem) & RWSEM_ACTIVE_MASK)
158 goto try_again_write;
162 * wait for a lock to be granted
164 static struct rw_semaphore __sched *
165 rwsem_down_failed_common(struct rw_semaphore *sem,
166 struct rwsem_waiter *waiter, signed long adjustment)
168 struct task_struct *tsk = current;
171 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
173 /* set up my own style of waitqueue */
174 spin_lock_irq(&sem->wait_lock);
176 get_task_struct(tsk);
178 list_add_tail(&waiter->list, &sem->wait_list);
180 /* we're now waiting on the lock, but no longer actively locking */
181 count = rwsem_atomic_update(adjustment, sem);
183 /* if there are no active locks, wake the front queued process(es) up */
184 if (!(count & RWSEM_ACTIVE_MASK))
185 sem = __rwsem_do_wake(sem, RWSEM_WAKE_NO_ACTIVE);
187 spin_unlock_irq(&sem->wait_lock);
189 /* wait to be given the lock */
194 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
197 tsk->state = TASK_RUNNING;
203 * wait for the read lock to be granted
205 asmregparm struct rw_semaphore __sched *
206 rwsem_down_read_failed(struct rw_semaphore *sem)
208 struct rwsem_waiter waiter;
210 waiter.flags = RWSEM_WAITING_FOR_READ;
211 rwsem_down_failed_common(sem, &waiter,
212 RWSEM_WAITING_BIAS - RWSEM_ACTIVE_BIAS);
217 * wait for the write lock to be granted
219 asmregparm struct rw_semaphore __sched *
220 rwsem_down_write_failed(struct rw_semaphore *sem)
222 struct rwsem_waiter waiter;
224 waiter.flags = RWSEM_WAITING_FOR_WRITE;
225 rwsem_down_failed_common(sem, &waiter, -RWSEM_ACTIVE_BIAS);
231 * handle waking up a waiter on the semaphore
232 * - up_read/up_write has decremented the active part of count if we come here
234 asmregparm struct rw_semaphore *rwsem_wake(struct rw_semaphore *sem)
238 spin_lock_irqsave(&sem->wait_lock, flags);
240 /* do nothing if list empty */
241 if (!list_empty(&sem->wait_list))
242 sem = __rwsem_do_wake(sem, RWSEM_WAKE_ANY);
244 spin_unlock_irqrestore(&sem->wait_lock, flags);
250 * downgrade a write lock into a read lock
251 * - caller incremented waiting part of count and discovered it still negative
252 * - just wake up any readers at the front of the queue
254 asmregparm struct rw_semaphore *rwsem_downgrade_wake(struct rw_semaphore *sem)
258 spin_lock_irqsave(&sem->wait_lock, flags);
260 /* do nothing if list empty */
261 if (!list_empty(&sem->wait_list))
262 sem = __rwsem_do_wake(sem, RWSEM_WAKE_READ_OWNED);
264 spin_unlock_irqrestore(&sem->wait_lock, flags);
269 EXPORT_SYMBOL(rwsem_down_read_failed);
270 EXPORT_SYMBOL(rwsem_down_write_failed);
271 EXPORT_SYMBOL(rwsem_wake);
272 EXPORT_SYMBOL(rwsem_downgrade_wake);