Root/lib/rwsem.c

1/* rwsem.c: R/W semaphores: contention handling functions
2 *
3 * Written by David Howells (dhowells@redhat.com).
4 * Derived from arch/i386/kernel/semaphore.c
5 */
6#include <linux/rwsem.h>
7#include <linux/sched.h>
8#include <linux/init.h>
9#include <linux/module.h>
10
11/*
12 * Initialize an rwsem:
13 */
14void __init_rwsem(struct rw_semaphore *sem, const char *name,
15          struct lock_class_key *key)
16{
17#ifdef CONFIG_DEBUG_LOCK_ALLOC
18    /*
19     * Make sure we are not reinitializing a held semaphore:
20     */
21    debug_check_no_locks_freed((void *)sem, sizeof(*sem));
22    lockdep_init_map(&sem->dep_map, name, key, 0);
23#endif
24    sem->count = RWSEM_UNLOCKED_VALUE;
25    spin_lock_init(&sem->wait_lock);
26    INIT_LIST_HEAD(&sem->wait_list);
27}
28
29EXPORT_SYMBOL(__init_rwsem);
30
31struct rwsem_waiter {
32    struct list_head list;
33    struct task_struct *task;
34    unsigned int flags;
35#define RWSEM_WAITING_FOR_READ 0x00000001
36#define RWSEM_WAITING_FOR_WRITE 0x00000002
37};
38
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.
42 */
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 */
46
47/*
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
56 */
57static struct rw_semaphore *
58__rwsem_do_wake(struct rw_semaphore *sem, int wake_type)
59{
60    struct rwsem_waiter *waiter;
61    struct task_struct *tsk;
62    struct list_head *next;
63    signed long oldcount, woken, loop, adjustment;
64
65    waiter = list_entry(sem->wait_list.next, struct rwsem_waiter, list);
66    if (!(waiter->flags & RWSEM_WAITING_FOR_WRITE))
67        goto readers_only;
68
69    if (wake_type == RWSEM_WAKE_READ_OWNED)
70        /* Another active reader was observed, so wakeup is not
71         * likely to succeed. Save the atomic op.
72         */
73        goto out;
74
75    /* There's a writer at the front of the queue - try to grant it the
76     * write lock. However, we only wake this writer if we can transition
77     * the active part of the count from 0 -> 1
78     */
79    adjustment = RWSEM_ACTIVE_WRITE_BIAS;
80    if (waiter->list.next == &sem->wait_list)
81        adjustment -= RWSEM_WAITING_BIAS;
82
83 try_again_write:
84    oldcount = rwsem_atomic_update(adjustment, sem) - adjustment;
85    if (oldcount & RWSEM_ACTIVE_MASK)
86        /* Someone grabbed the sem already */
87        goto undo_write;
88
89    /* We must be careful not to touch 'waiter' after we set ->task = NULL.
90     * It is an allocated on the waiter's stack and may become invalid at
91     * any time after that point (due to a wakeup from another source).
92     */
93    list_del(&waiter->list);
94    tsk = waiter->task;
95    smp_mb();
96    waiter->task = NULL;
97    wake_up_process(tsk);
98    put_task_struct(tsk);
99    goto out;
100
101 readers_only:
102    /* If we come here from up_xxxx(), another thread might have reached
103     * rwsem_down_failed_common() before we acquired the spinlock and
104     * woken up a waiter, making it now active. We prefer to check for
105     * this first in order to not spend too much time with the spinlock
106     * held if we're not going to be able to wake up readers in the end.
107     *
108     * Note that we do not need to update the rwsem count: any writer
109     * trying to acquire rwsem will run rwsem_down_write_failed() due
110     * to the waiting threads and block trying to acquire the spinlock.
111     *
112     * We use a dummy atomic update in order to acquire the cache line
113     * exclusively since we expect to succeed and run the final rwsem
114     * count adjustment pretty soon.
115     */
116    if (wake_type == RWSEM_WAKE_ANY &&
117        rwsem_atomic_update(0, sem) < RWSEM_WAITING_BIAS)
118        /* Someone grabbed the sem for write already */
119        goto out;
120
121    /* Grant an infinite number of read locks to the readers at the front
122     * of the queue. Note we increment the 'active part' of the count by
123     * the number of readers before waking any processes up.
124     */
125    woken = 0;
126    do {
127        woken++;
128
129        if (waiter->list.next == &sem->wait_list)
130            break;
131
132        waiter = list_entry(waiter->list.next,
133                    struct rwsem_waiter, list);
134
135    } while (waiter->flags & RWSEM_WAITING_FOR_READ);
136
137    adjustment = woken * RWSEM_ACTIVE_READ_BIAS;
138    if (waiter->flags & RWSEM_WAITING_FOR_READ)
139        /* hit end of list above */
140        adjustment -= RWSEM_WAITING_BIAS;
141
142    rwsem_atomic_add(adjustment, sem);
143
144    next = sem->wait_list.next;
145    for (loop = woken; loop > 0; loop--) {
146        waiter = list_entry(next, struct rwsem_waiter, list);
147        next = waiter->list.next;
148        tsk = waiter->task;
149        smp_mb();
150        waiter->task = NULL;
151        wake_up_process(tsk);
152        put_task_struct(tsk);
153    }
154
155    sem->wait_list.next = next;
156    next->prev = &sem->wait_list;
157
158 out:
159    return sem;
160
161    /* undo the change to the active count, but check for a transition
162     * 1->0 */
163 undo_write:
164    if (rwsem_atomic_update(-adjustment, sem) & RWSEM_ACTIVE_MASK)
165        goto out;
166    goto try_again_write;
167}
168
169/*
170 * wait for a lock to be granted
171 */
172static struct rw_semaphore __sched *
173rwsem_down_failed_common(struct rw_semaphore *sem,
174             unsigned int flags, signed long adjustment)
175{
176    struct rwsem_waiter waiter;
177    struct task_struct *tsk = current;
178    signed long count;
179
180    set_task_state(tsk, TASK_UNINTERRUPTIBLE);
181
182    /* set up my own style of waitqueue */
183    spin_lock_irq(&sem->wait_lock);
184    waiter.task = tsk;
185    waiter.flags = flags;
186    get_task_struct(tsk);
187
188    if (list_empty(&sem->wait_list))
189        adjustment += RWSEM_WAITING_BIAS;
190    list_add_tail(&waiter.list, &sem->wait_list);
191
192    /* we're now waiting on the lock, but no longer actively locking */
193    count = rwsem_atomic_update(adjustment, sem);
194
195    /* If there are no active locks, wake the front queued process(es) up.
196     *
197     * Alternatively, if we're called from a failed down_write(), there
198     * were already threads queued before us and there are no active
199     * writers, the lock must be read owned; so we try to wake any read
200     * locks that were queued ahead of us. */
201    if (count == RWSEM_WAITING_BIAS)
202        sem = __rwsem_do_wake(sem, RWSEM_WAKE_NO_ACTIVE);
203    else if (count > RWSEM_WAITING_BIAS &&
204         adjustment == -RWSEM_ACTIVE_WRITE_BIAS)
205        sem = __rwsem_do_wake(sem, RWSEM_WAKE_READ_OWNED);
206
207    spin_unlock_irq(&sem->wait_lock);
208
209    /* wait to be given the lock */
210    for (;;) {
211        if (!waiter.task)
212            break;
213        schedule();
214        set_task_state(tsk, TASK_UNINTERRUPTIBLE);
215    }
216
217    tsk->state = TASK_RUNNING;
218
219    return sem;
220}
221
222/*
223 * wait for the read lock to be granted
224 */
225asmregparm struct rw_semaphore __sched *
226rwsem_down_read_failed(struct rw_semaphore *sem)
227{
228    return rwsem_down_failed_common(sem, RWSEM_WAITING_FOR_READ,
229                    -RWSEM_ACTIVE_READ_BIAS);
230}
231
232/*
233 * wait for the write lock to be granted
234 */
235asmregparm struct rw_semaphore __sched *
236rwsem_down_write_failed(struct rw_semaphore *sem)
237{
238    return rwsem_down_failed_common(sem, RWSEM_WAITING_FOR_WRITE,
239                    -RWSEM_ACTIVE_WRITE_BIAS);
240}
241
242/*
243 * handle waking up a waiter on the semaphore
244 * - up_read/up_write has decremented the active part of count if we come here
245 */
246asmregparm struct rw_semaphore *rwsem_wake(struct rw_semaphore *sem)
247{
248    unsigned long flags;
249
250    spin_lock_irqsave(&sem->wait_lock, flags);
251
252    /* do nothing if list empty */
253    if (!list_empty(&sem->wait_list))
254        sem = __rwsem_do_wake(sem, RWSEM_WAKE_ANY);
255
256    spin_unlock_irqrestore(&sem->wait_lock, flags);
257
258    return sem;
259}
260
261/*
262 * downgrade a write lock into a read lock
263 * - caller incremented waiting part of count and discovered it still negative
264 * - just wake up any readers at the front of the queue
265 */
266asmregparm struct rw_semaphore *rwsem_downgrade_wake(struct rw_semaphore *sem)
267{
268    unsigned long flags;
269
270    spin_lock_irqsave(&sem->wait_lock, flags);
271
272    /* do nothing if list empty */
273    if (!list_empty(&sem->wait_list))
274        sem = __rwsem_do_wake(sem, RWSEM_WAKE_READ_OWNED);
275
276    spin_unlock_irqrestore(&sem->wait_lock, flags);
277
278    return sem;
279}
280
281EXPORT_SYMBOL(rwsem_down_read_failed);
282EXPORT_SYMBOL(rwsem_down_write_failed);
283EXPORT_SYMBOL(rwsem_wake);
284EXPORT_SYMBOL(rwsem_downgrade_wake);
285

Archive Download this file



interactive