xref: /linux/drivers/tty/tty_ldsem.c (revision ff5599816711d2e67da2d7561fd36ac48debd433)
1 /*
2  * Ldisc rw semaphore
3  *
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
9  *
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
14  *
15  * The generic counting was copied and modified from include/asm-generic/rwsem.h
16  * by Paul Mackerras <paulus@samba.org>.
17  *
18  * The scheduling policy was copied and modified from lib/rwsem.c
19  * Written by David Howells (dhowells@redhat.com).
20  *
21  * This implementation incorporates the write lock stealing work of
22  * Michel Lespinasse <walken@google.com>.
23  *
24  * Copyright (C) 2013 Peter Hurley <peter@hurleysoftware.com>
25  *
26  * This file may be redistributed under the terms of the GNU General Public
27  * License v2.
28  */
29 
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 
36 
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 # ifdef CONFIG_PROVE_LOCKING
43 #  define lockdep_acquire(l, s, t, i)		__acq(l, s, t, 0, 2, NULL, i)
44 #  define lockdep_acquire_nest(l, s, t, n, i)	__acq(l, s, t, 0, 2, n, i)
45 #  define lockdep_acquire_read(l, s, t, i)	__acq(l, s, t, 1, 2, NULL, i)
46 #  define lockdep_release(l, n, i)		__rel(l, n, i)
47 # else
48 #  define lockdep_acquire(l, s, t, i)		__acq(l, s, t, 0, 1, NULL, i)
49 #  define lockdep_acquire_nest(l, s, t, n, i)	__acq(l, s, t, 0, 1, n, i)
50 #  define lockdep_acquire_read(l, s, t, i)	__acq(l, s, t, 1, 1, NULL, i)
51 #  define lockdep_release(l, n, i)		__rel(l, n, i)
52 # endif
53 #else
54 # define lockdep_acquire(l, s, t, i)		do { } while (0)
55 # define lockdep_acquire_nest(l, s, t, n, i)	do { } while (0)
56 # define lockdep_acquire_read(l, s, t, i)	do { } while (0)
57 # define lockdep_release(l, n, i)		do { } while (0)
58 #endif
59 
60 #ifdef CONFIG_LOCK_STAT
61 # define lock_stat(_lock, stat)		lock_##stat(&(_lock)->dep_map, _RET_IP_)
62 #else
63 # define lock_stat(_lock, stat)		do { } while (0)
64 #endif
65 
66 
67 #if BITS_PER_LONG == 64
68 # define LDSEM_ACTIVE_MASK	0xffffffffL
69 #else
70 # define LDSEM_ACTIVE_MASK	0x0000ffffL
71 #endif
72 
73 #define LDSEM_UNLOCKED		0L
74 #define LDSEM_ACTIVE_BIAS	1L
75 #define LDSEM_WAIT_BIAS		(-LDSEM_ACTIVE_MASK-1)
76 #define LDSEM_READ_BIAS		LDSEM_ACTIVE_BIAS
77 #define LDSEM_WRITE_BIAS	(LDSEM_WAIT_BIAS + LDSEM_ACTIVE_BIAS)
78 
79 struct ldsem_waiter {
80 	struct list_head list;
81 	struct task_struct *task;
82 };
83 
84 static inline long ldsem_atomic_update(long delta, struct ld_semaphore *sem)
85 {
86 	return atomic_long_add_return(delta, (atomic_long_t *)&sem->count);
87 }
88 
89 static inline int ldsem_cmpxchg(long *old, long new, struct ld_semaphore *sem)
90 {
91 	long tmp = *old;
92 	*old = atomic_long_cmpxchg(&sem->count, *old, new);
93 	return *old == tmp;
94 }
95 
96 /*
97  * Initialize an ldsem:
98  */
99 void __init_ldsem(struct ld_semaphore *sem, const char *name,
100 		  struct lock_class_key *key)
101 {
102 #ifdef CONFIG_DEBUG_LOCK_ALLOC
103 	/*
104 	 * Make sure we are not reinitializing a held semaphore:
105 	 */
106 	debug_check_no_locks_freed((void *)sem, sizeof(*sem));
107 	lockdep_init_map(&sem->dep_map, name, key, 0);
108 #endif
109 	sem->count = LDSEM_UNLOCKED;
110 	sem->wait_readers = 0;
111 	raw_spin_lock_init(&sem->wait_lock);
112 	INIT_LIST_HEAD(&sem->read_wait);
113 	INIT_LIST_HEAD(&sem->write_wait);
114 }
115 
116 static void __ldsem_wake_readers(struct ld_semaphore *sem)
117 {
118 	struct ldsem_waiter *waiter, *next;
119 	struct task_struct *tsk;
120 	long adjust, count;
121 
122 	/* Try to grant read locks to all readers on the read wait list.
123 	 * Note the 'active part' of the count is incremented by
124 	 * the number of readers before waking any processes up.
125 	 */
126 	adjust = sem->wait_readers * (LDSEM_ACTIVE_BIAS - LDSEM_WAIT_BIAS);
127 	count = ldsem_atomic_update(adjust, sem);
128 	do {
129 		if (count > 0)
130 			break;
131 		if (ldsem_cmpxchg(&count, count - adjust, sem))
132 			return;
133 	} while (1);
134 
135 	list_for_each_entry_safe(waiter, next, &sem->read_wait, list) {
136 		tsk = waiter->task;
137 		smp_mb();
138 		waiter->task = NULL;
139 		wake_up_process(tsk);
140 		put_task_struct(tsk);
141 	}
142 	INIT_LIST_HEAD(&sem->read_wait);
143 	sem->wait_readers = 0;
144 }
145 
146 static inline int writer_trylock(struct ld_semaphore *sem)
147 {
148 	/* only wake this writer if the active part of the count can be
149 	 * transitioned from 0 -> 1
150 	 */
151 	long count = ldsem_atomic_update(LDSEM_ACTIVE_BIAS, sem);
152 	do {
153 		if ((count & LDSEM_ACTIVE_MASK) == LDSEM_ACTIVE_BIAS)
154 			return 1;
155 		if (ldsem_cmpxchg(&count, count - LDSEM_ACTIVE_BIAS, sem))
156 			return 0;
157 	} while (1);
158 }
159 
160 static void __ldsem_wake_writer(struct ld_semaphore *sem)
161 {
162 	struct ldsem_waiter *waiter;
163 
164 	waiter = list_entry(sem->write_wait.next, struct ldsem_waiter, list);
165 	wake_up_process(waiter->task);
166 }
167 
168 /*
169  * handle the lock release when processes blocked on it that can now run
170  * - if we come here from up_xxxx(), then:
171  *   - the 'active part' of count (&0x0000ffff) reached 0 (but may have changed)
172  *   - the 'waiting part' of count (&0xffff0000) is -ve (and will still be so)
173  * - the spinlock must be held by the caller
174  * - woken process blocks are discarded from the list after having task zeroed
175  */
176 static void __ldsem_wake(struct ld_semaphore *sem)
177 {
178 	if (!list_empty(&sem->write_wait))
179 		__ldsem_wake_writer(sem);
180 	else if (!list_empty(&sem->read_wait))
181 		__ldsem_wake_readers(sem);
182 }
183 
184 static void ldsem_wake(struct ld_semaphore *sem)
185 {
186 	unsigned long flags;
187 
188 	raw_spin_lock_irqsave(&sem->wait_lock, flags);
189 	__ldsem_wake(sem);
190 	raw_spin_unlock_irqrestore(&sem->wait_lock, flags);
191 }
192 
193 /*
194  * wait for the read lock to be granted
195  */
196 static struct ld_semaphore __sched *
197 down_read_failed(struct ld_semaphore *sem, long count, long timeout)
198 {
199 	struct ldsem_waiter waiter;
200 	struct task_struct *tsk = current;
201 	long adjust = -LDSEM_ACTIVE_BIAS + LDSEM_WAIT_BIAS;
202 
203 	/* set up my own style of waitqueue */
204 	raw_spin_lock_irq(&sem->wait_lock);
205 
206 	/* Try to reverse the lock attempt but if the count has changed
207 	 * so that reversing fails, check if there are are no waiters,
208 	 * and early-out if not */
209 	do {
210 		if (ldsem_cmpxchg(&count, count + adjust, sem))
211 			break;
212 		if (count > 0) {
213 			raw_spin_unlock_irq(&sem->wait_lock);
214 			return sem;
215 		}
216 	} while (1);
217 
218 	list_add_tail(&waiter.list, &sem->read_wait);
219 	sem->wait_readers++;
220 
221 	waiter.task = tsk;
222 	get_task_struct(tsk);
223 
224 	/* if there are no active locks, wake the new lock owner(s) */
225 	if ((count & LDSEM_ACTIVE_MASK) == 0)
226 		__ldsem_wake(sem);
227 
228 	raw_spin_unlock_irq(&sem->wait_lock);
229 
230 	/* wait to be given the lock */
231 	for (;;) {
232 		set_task_state(tsk, TASK_UNINTERRUPTIBLE);
233 
234 		if (!waiter.task)
235 			break;
236 		if (!timeout)
237 			break;
238 		timeout = schedule_timeout(timeout);
239 	}
240 
241 	__set_task_state(tsk, TASK_RUNNING);
242 
243 	if (!timeout) {
244 		/* lock timed out but check if this task was just
245 		 * granted lock ownership - if so, pretend there
246 		 * was no timeout; otherwise, cleanup lock wait */
247 		raw_spin_lock_irq(&sem->wait_lock);
248 		if (waiter.task) {
249 			ldsem_atomic_update(-LDSEM_WAIT_BIAS, sem);
250 			list_del(&waiter.list);
251 			raw_spin_unlock_irq(&sem->wait_lock);
252 			put_task_struct(waiter.task);
253 			return NULL;
254 		}
255 		raw_spin_unlock_irq(&sem->wait_lock);
256 	}
257 
258 	return sem;
259 }
260 
261 /*
262  * wait for the write lock to be granted
263  */
264 static struct ld_semaphore __sched *
265 down_write_failed(struct ld_semaphore *sem, long count, long timeout)
266 {
267 	struct ldsem_waiter waiter;
268 	struct task_struct *tsk = current;
269 	long adjust = -LDSEM_ACTIVE_BIAS;
270 	int locked = 0;
271 
272 	/* set up my own style of waitqueue */
273 	raw_spin_lock_irq(&sem->wait_lock);
274 
275 	/* Try to reverse the lock attempt but if the count has changed
276 	 * so that reversing fails, check if the lock is now owned,
277 	 * and early-out if so */
278 	do {
279 		if (ldsem_cmpxchg(&count, count + adjust, sem))
280 			break;
281 		if ((count & LDSEM_ACTIVE_MASK) == LDSEM_ACTIVE_BIAS) {
282 			raw_spin_unlock_irq(&sem->wait_lock);
283 			return sem;
284 		}
285 	} while (1);
286 
287 	list_add_tail(&waiter.list, &sem->write_wait);
288 
289 	waiter.task = tsk;
290 
291 	set_task_state(tsk, TASK_UNINTERRUPTIBLE);
292 	for (;;) {
293 		if (!timeout)
294 			break;
295 		raw_spin_unlock_irq(&sem->wait_lock);
296 		timeout = schedule_timeout(timeout);
297 		raw_spin_lock_irq(&sem->wait_lock);
298 		set_task_state(tsk, TASK_UNINTERRUPTIBLE);
299 		if ((locked = writer_trylock(sem)))
300 			break;
301 	}
302 
303 	if (!locked)
304 		ldsem_atomic_update(-LDSEM_WAIT_BIAS, sem);
305 	list_del(&waiter.list);
306 	raw_spin_unlock_irq(&sem->wait_lock);
307 
308 	__set_task_state(tsk, TASK_RUNNING);
309 
310 	/* lock wait may have timed out */
311 	if (!locked)
312 		return NULL;
313 	return sem;
314 }
315 
316 
317 
318 static inline int __ldsem_down_read_nested(struct ld_semaphore *sem,
319 					   int subclass, long timeout)
320 {
321 	long count;
322 
323 	lockdep_acquire_read(sem, subclass, 0, _RET_IP_);
324 
325 	count = ldsem_atomic_update(LDSEM_READ_BIAS, sem);
326 	if (count <= 0) {
327 		lock_stat(sem, contended);
328 		if (!down_read_failed(sem, count, timeout)) {
329 			lockdep_release(sem, 1, _RET_IP_);
330 			return 0;
331 		}
332 	}
333 	lock_stat(sem, acquired);
334 	return 1;
335 }
336 
337 static inline int __ldsem_down_write_nested(struct ld_semaphore *sem,
338 					    int subclass, long timeout)
339 {
340 	long count;
341 
342 	lockdep_acquire(sem, subclass, 0, _RET_IP_);
343 
344 	count = ldsem_atomic_update(LDSEM_WRITE_BIAS, sem);
345 	if ((count & LDSEM_ACTIVE_MASK) != LDSEM_ACTIVE_BIAS) {
346 		lock_stat(sem, contended);
347 		if (!down_write_failed(sem, count, timeout)) {
348 			lockdep_release(sem, 1, _RET_IP_);
349 			return 0;
350 		}
351 	}
352 	lock_stat(sem, acquired);
353 	return 1;
354 }
355 
356 
357 /*
358  * lock for reading -- returns 1 if successful, 0 if timed out
359  */
360 int __sched ldsem_down_read(struct ld_semaphore *sem, long timeout)
361 {
362 	might_sleep();
363 	return __ldsem_down_read_nested(sem, 0, timeout);
364 }
365 
366 /*
367  * trylock for reading -- returns 1 if successful, 0 if contention
368  */
369 int ldsem_down_read_trylock(struct ld_semaphore *sem)
370 {
371 	long count = sem->count;
372 
373 	while (count >= 0) {
374 		if (ldsem_cmpxchg(&count, count + LDSEM_READ_BIAS, sem)) {
375 			lockdep_acquire_read(sem, 0, 1, _RET_IP_);
376 			lock_stat(sem, acquired);
377 			return 1;
378 		}
379 	}
380 	return 0;
381 }
382 
383 /*
384  * lock for writing -- returns 1 if successful, 0 if timed out
385  */
386 int __sched ldsem_down_write(struct ld_semaphore *sem, long timeout)
387 {
388 	might_sleep();
389 	return __ldsem_down_write_nested(sem, 0, timeout);
390 }
391 
392 /*
393  * trylock for writing -- returns 1 if successful, 0 if contention
394  */
395 int ldsem_down_write_trylock(struct ld_semaphore *sem)
396 {
397 	long count = sem->count;
398 
399 	while ((count & LDSEM_ACTIVE_MASK) == 0) {
400 		if (ldsem_cmpxchg(&count, count + LDSEM_WRITE_BIAS, sem)) {
401 			lockdep_acquire(sem, 0, 1, _RET_IP_);
402 			lock_stat(sem, acquired);
403 			return 1;
404 		}
405 	}
406 	return 0;
407 }
408 
409 /*
410  * release a read lock
411  */
412 void ldsem_up_read(struct ld_semaphore *sem)
413 {
414 	long count;
415 
416 	lockdep_release(sem, 1, _RET_IP_);
417 
418 	count = ldsem_atomic_update(-LDSEM_READ_BIAS, sem);
419 	if (count < 0 && (count & LDSEM_ACTIVE_MASK) == 0)
420 		ldsem_wake(sem);
421 }
422 
423 /*
424  * release a write lock
425  */
426 void ldsem_up_write(struct ld_semaphore *sem)
427 {
428 	long count;
429 
430 	lockdep_release(sem, 1, _RET_IP_);
431 
432 	count = ldsem_atomic_update(-LDSEM_WRITE_BIAS, sem);
433 	if (count < 0)
434 		ldsem_wake(sem);
435 }
436 
437 
438 #ifdef CONFIG_DEBUG_LOCK_ALLOC
439 
440 int ldsem_down_read_nested(struct ld_semaphore *sem, int subclass, long timeout)
441 {
442 	might_sleep();
443 	return __ldsem_down_read_nested(sem, subclass, timeout);
444 }
445 
446 int ldsem_down_write_nested(struct ld_semaphore *sem, int subclass,
447 			    long timeout)
448 {
449 	might_sleep();
450 	return __ldsem_down_write_nested(sem, subclass, timeout);
451 }
452 
453 #endif
454