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
16 * by Paul Mackerras <paulus@samba.org>.
18 * The scheduling policy was copied and modified from lib/rwsem.c
19 * Written by David Howells (dhowells@redhat.com).
21 * This implementation incorporates the write lock stealing work of
22 * Michel Lespinasse <walken@google.com>.
24 * Copyright (C) 2013 Peter Hurley <peter@hurleysoftware.com>
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>
35 #include <linux/sched/debug.h>
36 #include <linux/sched/task.h>
39 #ifdef CONFIG_DEBUG_LOCK_ALLOC
40 # define __acq(l, s, t, r, c, n, i) \
41 lock_acquire(&(l)->dep_map, s, t, r, c, n, i)
42 # define __rel(l, n, i) \
43 lock_release(&(l)->dep_map, n, i)
44 #define lockdep_acquire(l, s, t, i) __acq(l, s, t, 0, 1, NULL, i)
45 #define lockdep_acquire_nest(l, s, t, n, i) __acq(l, s, t, 0, 1, n, i)
46 #define lockdep_acquire_read(l, s, t, i) __acq(l, s, t, 1, 1, NULL, i)
47 #define lockdep_release(l, n, i) __rel(l, n, i)
49 # define lockdep_acquire(l, s, t, i) do { } while (0)
50 # define lockdep_acquire_nest(l, s, t, n, i) do { } while (0)
51 # define lockdep_acquire_read(l, s, t, i) do { } while (0)
52 # define lockdep_release(l, n, i) do { } while (0)
55 #ifdef CONFIG_LOCK_STAT
56 # define lock_stat(_lock, stat) lock_##stat(&(_lock)->dep_map, _RET_IP_)
58 # define lock_stat(_lock, stat) do { } while (0)
62 #if BITS_PER_LONG == 64
63 # define LDSEM_ACTIVE_MASK 0xffffffffL
65 # define LDSEM_ACTIVE_MASK 0x0000ffffL
68 #define LDSEM_UNLOCKED 0L
69 #define LDSEM_ACTIVE_BIAS 1L
70 #define LDSEM_WAIT_BIAS (-LDSEM_ACTIVE_MASK-1)
71 #define LDSEM_READ_BIAS LDSEM_ACTIVE_BIAS
72 #define LDSEM_WRITE_BIAS (LDSEM_WAIT_BIAS + LDSEM_ACTIVE_BIAS)
75 struct list_head list;
76 struct task_struct *task;
79 static inline long ldsem_atomic_update(long delta, struct ld_semaphore *sem)
81 return atomic_long_add_return(delta, (atomic_long_t *)&sem->count);
85 * ldsem_cmpxchg() updates @*old with the last-known sem->count value.
86 * Returns 1 if count was successfully changed; @*old will have @new value.
87 * Returns 0 if count was not changed; @*old will have most recent sem->count
89 static inline int ldsem_cmpxchg(long *old, long new, struct ld_semaphore *sem)
91 long tmp = atomic_long_cmpxchg(&sem->count, *old, new);
102 * Initialize an ldsem:
104 void __init_ldsem(struct ld_semaphore *sem, const char *name,
105 struct lock_class_key *key)
107 #ifdef CONFIG_DEBUG_LOCK_ALLOC
109 * Make sure we are not reinitializing a held semaphore:
111 debug_check_no_locks_freed((void *)sem, sizeof(*sem));
112 lockdep_init_map(&sem->dep_map, name, key, 0);
114 sem->count = LDSEM_UNLOCKED;
115 sem->wait_readers = 0;
116 raw_spin_lock_init(&sem->wait_lock);
117 INIT_LIST_HEAD(&sem->read_wait);
118 INIT_LIST_HEAD(&sem->write_wait);
121 static void __ldsem_wake_readers(struct ld_semaphore *sem)
123 struct ldsem_waiter *waiter, *next;
124 struct task_struct *tsk;
127 /* Try to grant read locks to all readers on the read wait list.
128 * Note the 'active part' of the count is incremented by
129 * the number of readers before waking any processes up.
131 adjust = sem->wait_readers * (LDSEM_ACTIVE_BIAS - LDSEM_WAIT_BIAS);
132 count = ldsem_atomic_update(adjust, sem);
136 if (ldsem_cmpxchg(&count, count - adjust, sem))
140 list_for_each_entry_safe(waiter, next, &sem->read_wait, list) {
144 wake_up_process(tsk);
145 put_task_struct(tsk);
147 INIT_LIST_HEAD(&sem->read_wait);
148 sem->wait_readers = 0;
151 static inline int writer_trylock(struct ld_semaphore *sem)
153 /* only wake this writer if the active part of the count can be
154 * transitioned from 0 -> 1
156 long count = ldsem_atomic_update(LDSEM_ACTIVE_BIAS, sem);
158 if ((count & LDSEM_ACTIVE_MASK) == LDSEM_ACTIVE_BIAS)
160 if (ldsem_cmpxchg(&count, count - LDSEM_ACTIVE_BIAS, sem))
165 static void __ldsem_wake_writer(struct ld_semaphore *sem)
167 struct ldsem_waiter *waiter;
169 waiter = list_entry(sem->write_wait.next, struct ldsem_waiter, list);
170 wake_up_process(waiter->task);
174 * handle the lock release when processes blocked on it that can now run
175 * - if we come here from up_xxxx(), then:
176 * - the 'active part' of count (&0x0000ffff) reached 0 (but may have changed)
177 * - the 'waiting part' of count (&0xffff0000) is -ve (and will still be so)
178 * - the spinlock must be held by the caller
179 * - woken process blocks are discarded from the list after having task zeroed
181 static void __ldsem_wake(struct ld_semaphore *sem)
183 if (!list_empty(&sem->write_wait))
184 __ldsem_wake_writer(sem);
185 else if (!list_empty(&sem->read_wait))
186 __ldsem_wake_readers(sem);
189 static void ldsem_wake(struct ld_semaphore *sem)
193 raw_spin_lock_irqsave(&sem->wait_lock, flags);
195 raw_spin_unlock_irqrestore(&sem->wait_lock, flags);
199 * wait for the read lock to be granted
201 static struct ld_semaphore __sched *
202 down_read_failed(struct ld_semaphore *sem, long count, long timeout)
204 struct ldsem_waiter waiter;
205 long adjust = -LDSEM_ACTIVE_BIAS + LDSEM_WAIT_BIAS;
207 /* set up my own style of waitqueue */
208 raw_spin_lock_irq(&sem->wait_lock);
210 /* Try to reverse the lock attempt but if the count has changed
211 * so that reversing fails, check if there are are no waiters,
212 * and early-out if not */
214 if (ldsem_cmpxchg(&count, count + adjust, sem))
217 raw_spin_unlock_irq(&sem->wait_lock);
222 list_add_tail(&waiter.list, &sem->read_wait);
225 waiter.task = current;
226 get_task_struct(current);
228 /* if there are no active locks, wake the new lock owner(s) */
229 if ((count & LDSEM_ACTIVE_MASK) == 0)
232 raw_spin_unlock_irq(&sem->wait_lock);
234 /* wait to be given the lock */
236 set_current_state(TASK_UNINTERRUPTIBLE);
242 timeout = schedule_timeout(timeout);
245 __set_current_state(TASK_RUNNING);
248 /* lock timed out but check if this task was just
249 * granted lock ownership - if so, pretend there
250 * was no timeout; otherwise, cleanup lock wait */
251 raw_spin_lock_irq(&sem->wait_lock);
253 ldsem_atomic_update(-LDSEM_WAIT_BIAS, sem);
254 list_del(&waiter.list);
255 raw_spin_unlock_irq(&sem->wait_lock);
256 put_task_struct(waiter.task);
259 raw_spin_unlock_irq(&sem->wait_lock);
266 * wait for the write lock to be granted
268 static struct ld_semaphore __sched *
269 down_write_failed(struct ld_semaphore *sem, long count, long timeout)
271 struct ldsem_waiter waiter;
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);
292 waiter.task = current;
294 set_current_state(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_current_state(TASK_UNINTERRUPTIBLE);
302 locked = writer_trylock(sem);
308 ldsem_atomic_update(-LDSEM_WAIT_BIAS, sem);
309 list_del(&waiter.list);
312 * In case of timeout, wake up every reader who gave the right of way
313 * to writer. Prevent separation readers into two groups:
314 * one that helds semaphore and another that sleeps.
315 * (in case of no contention with a writer)
317 if (!locked && list_empty(&sem->write_wait))
318 __ldsem_wake_readers(sem);
320 raw_spin_unlock_irq(&sem->wait_lock);
322 __set_current_state(TASK_RUNNING);
324 /* lock wait may have timed out */
332 static int __ldsem_down_read_nested(struct ld_semaphore *sem,
333 int subclass, long timeout)
337 lockdep_acquire_read(sem, subclass, 0, _RET_IP_);
339 count = ldsem_atomic_update(LDSEM_READ_BIAS, sem);
341 lock_stat(sem, contended);
342 if (!down_read_failed(sem, count, timeout)) {
343 lockdep_release(sem, 1, _RET_IP_);
347 lock_stat(sem, acquired);
351 static int __ldsem_down_write_nested(struct ld_semaphore *sem,
352 int subclass, long timeout)
356 lockdep_acquire(sem, subclass, 0, _RET_IP_);
358 count = ldsem_atomic_update(LDSEM_WRITE_BIAS, sem);
359 if ((count & LDSEM_ACTIVE_MASK) != LDSEM_ACTIVE_BIAS) {
360 lock_stat(sem, contended);
361 if (!down_write_failed(sem, count, timeout)) {
362 lockdep_release(sem, 1, _RET_IP_);
366 lock_stat(sem, acquired);
372 * lock for reading -- returns 1 if successful, 0 if timed out
374 int __sched ldsem_down_read(struct ld_semaphore *sem, long timeout)
377 return __ldsem_down_read_nested(sem, 0, timeout);
381 * trylock for reading -- returns 1 if successful, 0 if contention
383 int ldsem_down_read_trylock(struct ld_semaphore *sem)
385 long count = sem->count;
388 if (ldsem_cmpxchg(&count, count + LDSEM_READ_BIAS, sem)) {
389 lockdep_acquire_read(sem, 0, 1, _RET_IP_);
390 lock_stat(sem, acquired);
398 * lock for writing -- returns 1 if successful, 0 if timed out
400 int __sched ldsem_down_write(struct ld_semaphore *sem, long timeout)
403 return __ldsem_down_write_nested(sem, 0, timeout);
407 * trylock for writing -- returns 1 if successful, 0 if contention
409 int ldsem_down_write_trylock(struct ld_semaphore *sem)
411 long count = sem->count;
413 while ((count & LDSEM_ACTIVE_MASK) == 0) {
414 if (ldsem_cmpxchg(&count, count + LDSEM_WRITE_BIAS, sem)) {
415 lockdep_acquire(sem, 0, 1, _RET_IP_);
416 lock_stat(sem, acquired);
424 * release a read lock
426 void ldsem_up_read(struct ld_semaphore *sem)
430 lockdep_release(sem, 1, _RET_IP_);
432 count = ldsem_atomic_update(-LDSEM_READ_BIAS, sem);
433 if (count < 0 && (count & LDSEM_ACTIVE_MASK) == 0)
438 * release a write lock
440 void ldsem_up_write(struct ld_semaphore *sem)
444 lockdep_release(sem, 1, _RET_IP_);
446 count = ldsem_atomic_update(-LDSEM_WRITE_BIAS, sem);
452 #ifdef CONFIG_DEBUG_LOCK_ALLOC
454 int ldsem_down_read_nested(struct ld_semaphore *sem, int subclass, long timeout)
457 return __ldsem_down_read_nested(sem, subclass, timeout);
460 int ldsem_down_write_nested(struct ld_semaphore *sem, int subclass,
464 return __ldsem_down_write_nested(sem, subclass, timeout);