xref: /linux/drivers/tty/tty_ldsem.c (revision e3b3d0f549c1d19b94e6ac55c66643166ea649ef)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Ldisc rw semaphore
4  *
5  * The ldisc semaphore is semantically a rw_semaphore but which enforces
6  * an alternate policy, namely:
7  *   1) Supports lock wait timeouts
8  *   2) Write waiter has priority
9  *   3) Downgrading is not supported
10  *
11  * Implementation notes:
12  *   1) Upper half of semaphore count is a wait count (differs from rwsem
13  *	in that rwsem normalizes the upper half to the wait bias)
14  *   2) Lacks overflow checking
15  *
16  * The generic counting was copied and modified from include/asm-generic/rwsem.h
17  * by Paul Mackerras <paulus@samba.org>.
18  *
19  * The scheduling policy was copied and modified from lib/rwsem.c
20  * Written by David Howells (dhowells@redhat.com).
21  *
22  * This implementation incorporates the write lock stealing work of
23  * Michel Lespinasse <walken@google.com>.
24  *
25  * Copyright (C) 2013 Peter Hurley <peter@hurleysoftware.com>
26  *
27  * This file may be redistributed under the terms of the GNU General Public
28  * License v2.
29  */
30 
31 #include <linux/list.h>
32 #include <linux/spinlock.h>
33 #include <linux/atomic.h>
34 #include <linux/tty.h>
35 #include <linux/sched.h>
36 #include <linux/sched/debug.h>
37 #include <linux/sched/task.h>
38 
39 
40 #ifdef CONFIG_DEBUG_LOCK_ALLOC
41 # define __acq(l, s, t, r, c, n, i)		\
42 				lock_acquire(&(l)->dep_map, s, t, r, c, n, i)
43 # define __rel(l, n, i)				\
44 				lock_release(&(l)->dep_map, n, i)
45 #define lockdep_acquire(l, s, t, i)		__acq(l, s, t, 0, 1, NULL, i)
46 #define lockdep_acquire_nest(l, s, t, n, i)	__acq(l, s, t, 0, 1, n, i)
47 #define lockdep_acquire_read(l, s, t, i)	__acq(l, s, t, 1, 1, NULL, i)
48 #define lockdep_release(l, n, i)		__rel(l, n, i)
49 #else
50 # define lockdep_acquire(l, s, t, i)		do { } while (0)
51 # define lockdep_acquire_nest(l, s, t, n, i)	do { } while (0)
52 # define lockdep_acquire_read(l, s, t, i)	do { } while (0)
53 # define lockdep_release(l, n, i)		do { } while (0)
54 #endif
55 
56 #ifdef CONFIG_LOCK_STAT
57 # define lock_stat(_lock, stat)		lock_##stat(&(_lock)->dep_map, _RET_IP_)
58 #else
59 # define lock_stat(_lock, stat)		do { } while (0)
60 #endif
61 
62 
63 #if BITS_PER_LONG == 64
64 # define LDSEM_ACTIVE_MASK	0xffffffffL
65 #else
66 # define LDSEM_ACTIVE_MASK	0x0000ffffL
67 #endif
68 
69 #define LDSEM_UNLOCKED		0L
70 #define LDSEM_ACTIVE_BIAS	1L
71 #define LDSEM_WAIT_BIAS		(-LDSEM_ACTIVE_MASK-1)
72 #define LDSEM_READ_BIAS		LDSEM_ACTIVE_BIAS
73 #define LDSEM_WRITE_BIAS	(LDSEM_WAIT_BIAS + LDSEM_ACTIVE_BIAS)
74 
75 struct ldsem_waiter {
76 	struct list_head list;
77 	struct task_struct *task;
78 };
79 
80 static inline long ldsem_atomic_update(long delta, struct ld_semaphore *sem)
81 {
82 	return atomic_long_add_return(delta, (atomic_long_t *)&sem->count);
83 }
84 
85 /*
86  * ldsem_cmpxchg() updates @*old with the last-known sem->count value.
87  * Returns 1 if count was successfully changed; @*old will have @new value.
88  * Returns 0 if count was not changed; @*old will have most recent sem->count
89  */
90 static inline int ldsem_cmpxchg(long *old, long new, struct ld_semaphore *sem)
91 {
92 	long tmp = atomic_long_cmpxchg(&sem->count, *old, new);
93 	if (tmp == *old) {
94 		*old = new;
95 		return 1;
96 	} else {
97 		*old = tmp;
98 		return 0;
99 	}
100 }
101 
102 /*
103  * Initialize an ldsem:
104  */
105 void __init_ldsem(struct ld_semaphore *sem, const char *name,
106 		  struct lock_class_key *key)
107 {
108 #ifdef CONFIG_DEBUG_LOCK_ALLOC
109 	/*
110 	 * Make sure we are not reinitializing a held semaphore:
111 	 */
112 	debug_check_no_locks_freed((void *)sem, sizeof(*sem));
113 	lockdep_init_map(&sem->dep_map, name, key, 0);
114 #endif
115 	sem->count = LDSEM_UNLOCKED;
116 	sem->wait_readers = 0;
117 	raw_spin_lock_init(&sem->wait_lock);
118 	INIT_LIST_HEAD(&sem->read_wait);
119 	INIT_LIST_HEAD(&sem->write_wait);
120 }
121 
122 static void __ldsem_wake_readers(struct ld_semaphore *sem)
123 {
124 	struct ldsem_waiter *waiter, *next;
125 	struct task_struct *tsk;
126 	long adjust, count;
127 
128 	/* Try to grant read locks to all readers on the read wait list.
129 	 * Note the 'active part' of the count is incremented by
130 	 * the number of readers before waking any processes up.
131 	 */
132 	adjust = sem->wait_readers * (LDSEM_ACTIVE_BIAS - LDSEM_WAIT_BIAS);
133 	count = ldsem_atomic_update(adjust, sem);
134 	do {
135 		if (count > 0)
136 			break;
137 		if (ldsem_cmpxchg(&count, count - adjust, sem))
138 			return;
139 	} while (1);
140 
141 	list_for_each_entry_safe(waiter, next, &sem->read_wait, list) {
142 		tsk = waiter->task;
143 		smp_mb();
144 		waiter->task = NULL;
145 		wake_up_process(tsk);
146 		put_task_struct(tsk);
147 	}
148 	INIT_LIST_HEAD(&sem->read_wait);
149 	sem->wait_readers = 0;
150 }
151 
152 static inline int writer_trylock(struct ld_semaphore *sem)
153 {
154 	/* only wake this writer if the active part of the count can be
155 	 * transitioned from 0 -> 1
156 	 */
157 	long count = ldsem_atomic_update(LDSEM_ACTIVE_BIAS, sem);
158 	do {
159 		if ((count & LDSEM_ACTIVE_MASK) == LDSEM_ACTIVE_BIAS)
160 			return 1;
161 		if (ldsem_cmpxchg(&count, count - LDSEM_ACTIVE_BIAS, sem))
162 			return 0;
163 	} while (1);
164 }
165 
166 static void __ldsem_wake_writer(struct ld_semaphore *sem)
167 {
168 	struct ldsem_waiter *waiter;
169 
170 	waiter = list_entry(sem->write_wait.next, struct ldsem_waiter, list);
171 	wake_up_process(waiter->task);
172 }
173 
174 /*
175  * handle the lock release when processes blocked on it that can now run
176  * - if we come here from up_xxxx(), then:
177  *   - the 'active part' of count (&0x0000ffff) reached 0 (but may have changed)
178  *   - the 'waiting part' of count (&0xffff0000) is -ve (and will still be so)
179  * - the spinlock must be held by the caller
180  * - woken process blocks are discarded from the list after having task zeroed
181  */
182 static void __ldsem_wake(struct ld_semaphore *sem)
183 {
184 	if (!list_empty(&sem->write_wait))
185 		__ldsem_wake_writer(sem);
186 	else if (!list_empty(&sem->read_wait))
187 		__ldsem_wake_readers(sem);
188 }
189 
190 static void ldsem_wake(struct ld_semaphore *sem)
191 {
192 	unsigned long flags;
193 
194 	raw_spin_lock_irqsave(&sem->wait_lock, flags);
195 	__ldsem_wake(sem);
196 	raw_spin_unlock_irqrestore(&sem->wait_lock, flags);
197 }
198 
199 /*
200  * wait for the read lock to be granted
201  */
202 static struct ld_semaphore __sched *
203 down_read_failed(struct ld_semaphore *sem, long count, long timeout)
204 {
205 	struct ldsem_waiter waiter;
206 	long adjust = -LDSEM_ACTIVE_BIAS + LDSEM_WAIT_BIAS;
207 
208 	/* set up my own style of waitqueue */
209 	raw_spin_lock_irq(&sem->wait_lock);
210 
211 	/* Try to reverse the lock attempt but if the count has changed
212 	 * so that reversing fails, check if there are are no waiters,
213 	 * and early-out if not */
214 	do {
215 		if (ldsem_cmpxchg(&count, count + adjust, sem))
216 			break;
217 		if (count > 0) {
218 			raw_spin_unlock_irq(&sem->wait_lock);
219 			return sem;
220 		}
221 	} while (1);
222 
223 	list_add_tail(&waiter.list, &sem->read_wait);
224 	sem->wait_readers++;
225 
226 	waiter.task = current;
227 	get_task_struct(current);
228 
229 	/* if there are no active locks, wake the new lock owner(s) */
230 	if ((count & LDSEM_ACTIVE_MASK) == 0)
231 		__ldsem_wake(sem);
232 
233 	raw_spin_unlock_irq(&sem->wait_lock);
234 
235 	/* wait to be given the lock */
236 	for (;;) {
237 		set_current_state(TASK_UNINTERRUPTIBLE);
238 
239 		if (!waiter.task)
240 			break;
241 		if (!timeout)
242 			break;
243 		timeout = schedule_timeout(timeout);
244 	}
245 
246 	__set_current_state(TASK_RUNNING);
247 
248 	if (!timeout) {
249 		/* lock timed out but check if this task was just
250 		 * granted lock ownership - if so, pretend there
251 		 * was no timeout; otherwise, cleanup lock wait */
252 		raw_spin_lock_irq(&sem->wait_lock);
253 		if (waiter.task) {
254 			ldsem_atomic_update(-LDSEM_WAIT_BIAS, sem);
255 			list_del(&waiter.list);
256 			raw_spin_unlock_irq(&sem->wait_lock);
257 			put_task_struct(waiter.task);
258 			return NULL;
259 		}
260 		raw_spin_unlock_irq(&sem->wait_lock);
261 	}
262 
263 	return sem;
264 }
265 
266 /*
267  * wait for the write lock to be granted
268  */
269 static struct ld_semaphore __sched *
270 down_write_failed(struct ld_semaphore *sem, long count, long timeout)
271 {
272 	struct ldsem_waiter waiter;
273 	long adjust = -LDSEM_ACTIVE_BIAS;
274 	int locked = 0;
275 
276 	/* set up my own style of waitqueue */
277 	raw_spin_lock_irq(&sem->wait_lock);
278 
279 	/* Try to reverse the lock attempt but if the count has changed
280 	 * so that reversing fails, check if the lock is now owned,
281 	 * and early-out if so */
282 	do {
283 		if (ldsem_cmpxchg(&count, count + adjust, sem))
284 			break;
285 		if ((count & LDSEM_ACTIVE_MASK) == LDSEM_ACTIVE_BIAS) {
286 			raw_spin_unlock_irq(&sem->wait_lock);
287 			return sem;
288 		}
289 	} while (1);
290 
291 	list_add_tail(&waiter.list, &sem->write_wait);
292 
293 	waiter.task = current;
294 
295 	set_current_state(TASK_UNINTERRUPTIBLE);
296 	for (;;) {
297 		if (!timeout)
298 			break;
299 		raw_spin_unlock_irq(&sem->wait_lock);
300 		timeout = schedule_timeout(timeout);
301 		raw_spin_lock_irq(&sem->wait_lock);
302 		set_current_state(TASK_UNINTERRUPTIBLE);
303 		locked = writer_trylock(sem);
304 		if (locked)
305 			break;
306 	}
307 
308 	if (!locked)
309 		ldsem_atomic_update(-LDSEM_WAIT_BIAS, sem);
310 	list_del(&waiter.list);
311 	raw_spin_unlock_irq(&sem->wait_lock);
312 
313 	__set_current_state(TASK_RUNNING);
314 
315 	/* lock wait may have timed out */
316 	if (!locked)
317 		return NULL;
318 	return sem;
319 }
320 
321 
322 
323 static int __ldsem_down_read_nested(struct ld_semaphore *sem,
324 					   int subclass, long timeout)
325 {
326 	long count;
327 
328 	lockdep_acquire_read(sem, subclass, 0, _RET_IP_);
329 
330 	count = ldsem_atomic_update(LDSEM_READ_BIAS, sem);
331 	if (count <= 0) {
332 		lock_stat(sem, contended);
333 		if (!down_read_failed(sem, count, timeout)) {
334 			lockdep_release(sem, 1, _RET_IP_);
335 			return 0;
336 		}
337 	}
338 	lock_stat(sem, acquired);
339 	return 1;
340 }
341 
342 static int __ldsem_down_write_nested(struct ld_semaphore *sem,
343 					    int subclass, long timeout)
344 {
345 	long count;
346 
347 	lockdep_acquire(sem, subclass, 0, _RET_IP_);
348 
349 	count = ldsem_atomic_update(LDSEM_WRITE_BIAS, sem);
350 	if ((count & LDSEM_ACTIVE_MASK) != LDSEM_ACTIVE_BIAS) {
351 		lock_stat(sem, contended);
352 		if (!down_write_failed(sem, count, timeout)) {
353 			lockdep_release(sem, 1, _RET_IP_);
354 			return 0;
355 		}
356 	}
357 	lock_stat(sem, acquired);
358 	return 1;
359 }
360 
361 
362 /*
363  * lock for reading -- returns 1 if successful, 0 if timed out
364  */
365 int __sched ldsem_down_read(struct ld_semaphore *sem, long timeout)
366 {
367 	might_sleep();
368 	return __ldsem_down_read_nested(sem, 0, timeout);
369 }
370 
371 /*
372  * trylock for reading -- returns 1 if successful, 0 if contention
373  */
374 int ldsem_down_read_trylock(struct ld_semaphore *sem)
375 {
376 	long count = sem->count;
377 
378 	while (count >= 0) {
379 		if (ldsem_cmpxchg(&count, count + LDSEM_READ_BIAS, sem)) {
380 			lockdep_acquire_read(sem, 0, 1, _RET_IP_);
381 			lock_stat(sem, acquired);
382 			return 1;
383 		}
384 	}
385 	return 0;
386 }
387 
388 /*
389  * lock for writing -- returns 1 if successful, 0 if timed out
390  */
391 int __sched ldsem_down_write(struct ld_semaphore *sem, long timeout)
392 {
393 	might_sleep();
394 	return __ldsem_down_write_nested(sem, 0, timeout);
395 }
396 
397 /*
398  * trylock for writing -- returns 1 if successful, 0 if contention
399  */
400 int ldsem_down_write_trylock(struct ld_semaphore *sem)
401 {
402 	long count = sem->count;
403 
404 	while ((count & LDSEM_ACTIVE_MASK) == 0) {
405 		if (ldsem_cmpxchg(&count, count + LDSEM_WRITE_BIAS, sem)) {
406 			lockdep_acquire(sem, 0, 1, _RET_IP_);
407 			lock_stat(sem, acquired);
408 			return 1;
409 		}
410 	}
411 	return 0;
412 }
413 
414 /*
415  * release a read lock
416  */
417 void ldsem_up_read(struct ld_semaphore *sem)
418 {
419 	long count;
420 
421 	lockdep_release(sem, 1, _RET_IP_);
422 
423 	count = ldsem_atomic_update(-LDSEM_READ_BIAS, sem);
424 	if (count < 0 && (count & LDSEM_ACTIVE_MASK) == 0)
425 		ldsem_wake(sem);
426 }
427 
428 /*
429  * release a write lock
430  */
431 void ldsem_up_write(struct ld_semaphore *sem)
432 {
433 	long count;
434 
435 	lockdep_release(sem, 1, _RET_IP_);
436 
437 	count = ldsem_atomic_update(-LDSEM_WRITE_BIAS, sem);
438 	if (count < 0)
439 		ldsem_wake(sem);
440 }
441 
442 
443 #ifdef CONFIG_DEBUG_LOCK_ALLOC
444 
445 int ldsem_down_read_nested(struct ld_semaphore *sem, int subclass, long timeout)
446 {
447 	might_sleep();
448 	return __ldsem_down_read_nested(sem, subclass, timeout);
449 }
450 
451 int ldsem_down_write_nested(struct ld_semaphore *sem, int subclass,
452 			    long timeout)
453 {
454 	might_sleep();
455 	return __ldsem_down_write_nested(sem, subclass, timeout);
456 }
457 
458 #endif
459