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 struct list_head list;
13 struct task_struct *task;
15 #define RWSEM_WAITING_FOR_READ 0x00000001
16 #define RWSEM_WAITING_FOR_WRITE 0x00000002
21 void rwsemtrace(struct rw_semaphore *sem, const char *str)
23 printk("sem=%p\n", sem);
24 printk("(sem)=%08lx\n", sem->count);
26 printk("[%d] %s({%08lx})\n", current->pid, str, sem->count);
31 * handle the lock release when processes blocked on it that can now run
32 * - if we come here from up_xxxx(), then:
33 * - the 'active part' of count (&0x0000ffff) reached 0 (but may have changed)
34 * - the 'waiting part' of count (&0xffff0000) is -ve (and will still be so)
35 * - there must be someone on the queue
36 * - the spinlock must be held by the caller
37 * - woken process blocks are discarded from the list after having task zeroed
38 * - writers are only woken if downgrading is false
40 static inline struct rw_semaphore *
41 __rwsem_do_wake(struct rw_semaphore *sem, int downgrading)
43 struct rwsem_waiter *waiter;
44 struct task_struct *tsk;
45 struct list_head *next;
46 signed long oldcount, woken, loop;
48 rwsemtrace(sem, "Entering __rwsem_do_wake");
51 goto dont_wake_writers;
53 /* if we came through an up_xxxx() call, we only only wake someone up
54 * if we can transition the active part of the count from 0 -> 1
57 oldcount = rwsem_atomic_update(RWSEM_ACTIVE_BIAS, sem)
59 if (oldcount & RWSEM_ACTIVE_MASK)
62 waiter = list_entry(sem->wait_list.next, struct rwsem_waiter, list);
64 /* try to grant a single write lock if there's a writer at the front
65 * of the queue - note we leave the 'active part' of the count
66 * incremented by 1 and the waiting part incremented by 0x00010000
68 if (!(waiter->flags & RWSEM_WAITING_FOR_WRITE))
71 /* We must be careful not to touch 'waiter' after we set ->task = NULL.
72 * It is an allocated on the waiter's stack and may become invalid at
73 * any time after that point (due to a wakeup from another source).
75 list_del(&waiter->list);
83 /* don't want to wake any writers */
85 waiter = list_entry(sem->wait_list.next, struct rwsem_waiter, list);
86 if (waiter->flags & RWSEM_WAITING_FOR_WRITE)
89 /* grant an infinite number of read locks to the readers at the front
91 * - note we increment the 'active part' of the count by the number of
92 * readers before waking any processes up
99 if (waiter->list.next == &sem->wait_list)
102 waiter = list_entry(waiter->list.next,
103 struct rwsem_waiter, list);
105 } while (waiter->flags & RWSEM_WAITING_FOR_READ);
108 woken *= RWSEM_ACTIVE_BIAS - RWSEM_WAITING_BIAS;
110 /* we'd already done one increment earlier */
111 woken -= RWSEM_ACTIVE_BIAS;
113 rwsem_atomic_add(woken, sem);
115 next = sem->wait_list.next;
116 for (; loop > 0; loop--) {
117 waiter = list_entry(next, struct rwsem_waiter, list);
118 next = waiter->list.next;
122 wake_up_process(tsk);
123 put_task_struct(tsk);
126 sem->wait_list.next = next;
127 next->prev = &sem->wait_list;
130 rwsemtrace(sem, "Leaving __rwsem_do_wake");
133 /* undo the change to count, but check for a transition 1->0 */
135 if (rwsem_atomic_update(-RWSEM_ACTIVE_BIAS, sem) != 0)
141 * wait for a lock to be granted
143 static inline struct rw_semaphore *
144 rwsem_down_failed_common(struct rw_semaphore *sem,
145 struct rwsem_waiter *waiter, signed long adjustment)
147 struct task_struct *tsk = current;
150 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
152 /* set up my own style of waitqueue */
153 spin_lock_irq(&sem->wait_lock);
155 get_task_struct(tsk);
157 list_add_tail(&waiter->list, &sem->wait_list);
159 /* we're now waiting on the lock, but no longer actively read-locking */
160 count = rwsem_atomic_update(adjustment, sem);
162 /* if there are no active locks, wake the front queued process(es) up */
163 if (!(count & RWSEM_ACTIVE_MASK))
164 sem = __rwsem_do_wake(sem, 0);
166 spin_unlock_irq(&sem->wait_lock);
168 /* wait to be given the lock */
173 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
176 tsk->state = TASK_RUNNING;
182 * wait for the read lock to be granted
184 struct rw_semaphore fastcall __sched *
185 rwsem_down_read_failed(struct rw_semaphore *sem)
187 struct rwsem_waiter waiter;
189 rwsemtrace(sem, "Entering rwsem_down_read_failed");
191 waiter.flags = RWSEM_WAITING_FOR_READ;
192 rwsem_down_failed_common(sem, &waiter,
193 RWSEM_WAITING_BIAS - RWSEM_ACTIVE_BIAS);
195 rwsemtrace(sem, "Leaving rwsem_down_read_failed");
200 * wait for the write lock to be granted
202 struct rw_semaphore fastcall __sched *
203 rwsem_down_write_failed(struct rw_semaphore *sem)
205 struct rwsem_waiter waiter;
207 rwsemtrace(sem, "Entering rwsem_down_write_failed");
209 waiter.flags = RWSEM_WAITING_FOR_WRITE;
210 rwsem_down_failed_common(sem, &waiter, -RWSEM_ACTIVE_BIAS);
212 rwsemtrace(sem, "Leaving rwsem_down_write_failed");
217 * handle waking up a waiter on the semaphore
218 * - up_read/up_write has decremented the active part of count if we come here
220 struct rw_semaphore fastcall *rwsem_wake(struct rw_semaphore *sem)
224 rwsemtrace(sem, "Entering rwsem_wake");
226 spin_lock_irqsave(&sem->wait_lock, flags);
228 /* do nothing if list empty */
229 if (!list_empty(&sem->wait_list))
230 sem = __rwsem_do_wake(sem, 0);
232 spin_unlock_irqrestore(&sem->wait_lock, flags);
234 rwsemtrace(sem, "Leaving rwsem_wake");
240 * downgrade a write lock into a read lock
241 * - caller incremented waiting part of count and discovered it still negative
242 * - just wake up any readers at the front of the queue
244 struct rw_semaphore fastcall *rwsem_downgrade_wake(struct rw_semaphore *sem)
248 rwsemtrace(sem, "Entering rwsem_downgrade_wake");
250 spin_lock_irqsave(&sem->wait_lock, flags);
252 /* do nothing if list empty */
253 if (!list_empty(&sem->wait_list))
254 sem = __rwsem_do_wake(sem, 1);
256 spin_unlock_irqrestore(&sem->wait_lock, flags);
258 rwsemtrace(sem, "Leaving rwsem_downgrade_wake");
262 EXPORT_SYMBOL(rwsem_down_read_failed);
263 EXPORT_SYMBOL(rwsem_down_write_failed);
264 EXPORT_SYMBOL(rwsem_wake);
265 EXPORT_SYMBOL(rwsem_downgrade_wake);
267 EXPORT_SYMBOL(rwsemtrace);