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 
28 #include <linux/list.h>
29 #include <linux/spinlock.h>
30 #include <linux/atomic.h>
31 #include <linux/tty.h>
32 #include <linux/sched.h>
33 #include <linux/sched/debug.h>
34 #include <linux/sched/task.h>
35 
36 
37 #if BITS_PER_LONG == 64
38 # define LDSEM_ACTIVE_MASK	0xffffffffL
39 #else
40 # define LDSEM_ACTIVE_MASK	0x0000ffffL
41 #endif
42 
43 #define LDSEM_UNLOCKED		0L
44 #define LDSEM_ACTIVE_BIAS	1L
45 #define LDSEM_WAIT_BIAS		(-LDSEM_ACTIVE_MASK-1)
46 #define LDSEM_READ_BIAS		LDSEM_ACTIVE_BIAS
47 #define LDSEM_WRITE_BIAS	(LDSEM_WAIT_BIAS + LDSEM_ACTIVE_BIAS)
48 
49 struct ldsem_waiter {
50 	struct list_head list;
51 	struct task_struct *task;
52 };
53 
54 /*
55  * Initialize an ldsem:
56  */
__init_ldsem(struct ld_semaphore * sem,const char * name,struct lock_class_key * key)57 void __init_ldsem(struct ld_semaphore *sem, const char *name,
58 		  struct lock_class_key *key)
59 {
60 #ifdef CONFIG_DEBUG_LOCK_ALLOC
61 	/*
62 	 * Make sure we are not reinitializing a held semaphore:
63 	 */
64 	debug_check_no_locks_freed((void *)sem, sizeof(*sem));
65 	lockdep_init_map(&sem->dep_map, name, key, 0);
66 #endif
67 	atomic_long_set(&sem->count, LDSEM_UNLOCKED);
68 	sem->wait_readers = 0;
69 	raw_spin_lock_init(&sem->wait_lock);
70 	INIT_LIST_HEAD(&sem->read_wait);
71 	INIT_LIST_HEAD(&sem->write_wait);
72 }
73 
__ldsem_wake_readers(struct ld_semaphore * sem)74 static void __ldsem_wake_readers(struct ld_semaphore *sem)
75 {
76 	struct ldsem_waiter *waiter, *next;
77 	struct task_struct *tsk;
78 	long adjust, count;
79 
80 	/*
81 	 * Try to grant read locks to all readers on the read wait list.
82 	 * Note the 'active part' of the count is incremented by
83 	 * the number of readers before waking any processes up.
84 	 */
85 	adjust = sem->wait_readers * (LDSEM_ACTIVE_BIAS - LDSEM_WAIT_BIAS);
86 	count = atomic_long_add_return(adjust, &sem->count);
87 	do {
88 		if (count > 0)
89 			break;
90 		if (atomic_long_try_cmpxchg(&sem->count, &count, count - adjust))
91 			return;
92 	} while (1);
93 
94 	list_for_each_entry_safe(waiter, next, &sem->read_wait, list) {
95 		tsk = waiter->task;
96 		smp_store_release(&waiter->task, NULL);
97 		wake_up_process(tsk);
98 		put_task_struct(tsk);
99 	}
100 	INIT_LIST_HEAD(&sem->read_wait);
101 	sem->wait_readers = 0;
102 }
103 
writer_trylock(struct ld_semaphore * sem)104 static inline int writer_trylock(struct ld_semaphore *sem)
105 {
106 	/*
107 	 * Only wake this writer if the active part of the count can be
108 	 * transitioned from 0 -> 1
109 	 */
110 	long count = atomic_long_add_return(LDSEM_ACTIVE_BIAS, &sem->count);
111 	do {
112 		if ((count & LDSEM_ACTIVE_MASK) == LDSEM_ACTIVE_BIAS)
113 			return 1;
114 		if (atomic_long_try_cmpxchg(&sem->count, &count, count - LDSEM_ACTIVE_BIAS))
115 			return 0;
116 	} while (1);
117 }
118 
__ldsem_wake_writer(struct ld_semaphore * sem)119 static void __ldsem_wake_writer(struct ld_semaphore *sem)
120 {
121 	struct ldsem_waiter *waiter;
122 
123 	waiter = list_entry(sem->write_wait.next, struct ldsem_waiter, list);
124 	wake_up_process(waiter->task);
125 }
126 
127 /*
128  * handle the lock release when processes blocked on it that can now run
129  * - if we come here from up_xxxx(), then:
130  *   - the 'active part' of count (&0x0000ffff) reached 0 (but may have changed)
131  *   - the 'waiting part' of count (&0xffff0000) is -ve (and will still be so)
132  * - the spinlock must be held by the caller
133  * - woken process blocks are discarded from the list after having task zeroed
134  */
__ldsem_wake(struct ld_semaphore * sem)135 static void __ldsem_wake(struct ld_semaphore *sem)
136 {
137 	if (!list_empty(&sem->write_wait))
138 		__ldsem_wake_writer(sem);
139 	else if (!list_empty(&sem->read_wait))
140 		__ldsem_wake_readers(sem);
141 }
142 
ldsem_wake(struct ld_semaphore * sem)143 static void ldsem_wake(struct ld_semaphore *sem)
144 {
145 	unsigned long flags;
146 
147 	raw_spin_lock_irqsave(&sem->wait_lock, flags);
148 	__ldsem_wake(sem);
149 	raw_spin_unlock_irqrestore(&sem->wait_lock, flags);
150 }
151 
152 /*
153  * wait for the read lock to be granted
154  */
155 static struct ld_semaphore __sched *
down_read_failed(struct ld_semaphore * sem,long count,long timeout)156 down_read_failed(struct ld_semaphore *sem, long count, long timeout)
157 {
158 	struct ldsem_waiter waiter;
159 	long adjust = -LDSEM_ACTIVE_BIAS + LDSEM_WAIT_BIAS;
160 
161 	/* set up my own style of waitqueue */
162 	raw_spin_lock_irq(&sem->wait_lock);
163 
164 	/*
165 	 * Try to reverse the lock attempt but if the count has changed
166 	 * so that reversing fails, check if there are no waiters,
167 	 * and early-out if not
168 	 */
169 	do {
170 		if (atomic_long_try_cmpxchg(&sem->count, &count, count + adjust)) {
171 			count += adjust;
172 			break;
173 		}
174 		if (count > 0) {
175 			raw_spin_unlock_irq(&sem->wait_lock);
176 			return sem;
177 		}
178 	} while (1);
179 
180 	list_add_tail(&waiter.list, &sem->read_wait);
181 	sem->wait_readers++;
182 
183 	waiter.task = current;
184 	get_task_struct(current);
185 
186 	/* if there are no active locks, wake the new lock owner(s) */
187 	if ((count & LDSEM_ACTIVE_MASK) == 0)
188 		__ldsem_wake(sem);
189 
190 	raw_spin_unlock_irq(&sem->wait_lock);
191 
192 	/* wait to be given the lock */
193 	for (;;) {
194 		set_current_state(TASK_UNINTERRUPTIBLE);
195 
196 		if (!smp_load_acquire(&waiter.task))
197 			break;
198 		if (!timeout)
199 			break;
200 		timeout = schedule_timeout(timeout);
201 	}
202 
203 	__set_current_state(TASK_RUNNING);
204 
205 	if (!timeout) {
206 		/*
207 		 * Lock timed out but check if this task was just
208 		 * granted lock ownership - if so, pretend there
209 		 * was no timeout; otherwise, cleanup lock wait.
210 		 */
211 		raw_spin_lock_irq(&sem->wait_lock);
212 		if (waiter.task) {
213 			atomic_long_add_return(-LDSEM_WAIT_BIAS, &sem->count);
214 			sem->wait_readers--;
215 			list_del(&waiter.list);
216 			raw_spin_unlock_irq(&sem->wait_lock);
217 			put_task_struct(waiter.task);
218 			return NULL;
219 		}
220 		raw_spin_unlock_irq(&sem->wait_lock);
221 	}
222 
223 	return sem;
224 }
225 
226 /*
227  * wait for the write lock to be granted
228  */
229 static struct ld_semaphore __sched *
down_write_failed(struct ld_semaphore * sem,long count,long timeout)230 down_write_failed(struct ld_semaphore *sem, long count, long timeout)
231 {
232 	struct ldsem_waiter waiter;
233 	long adjust = -LDSEM_ACTIVE_BIAS;
234 	int locked = 0;
235 
236 	/* set up my own style of waitqueue */
237 	raw_spin_lock_irq(&sem->wait_lock);
238 
239 	/*
240 	 * Try to reverse the lock attempt but if the count has changed
241 	 * so that reversing fails, check if the lock is now owned,
242 	 * and early-out if so.
243 	 */
244 	do {
245 		if (atomic_long_try_cmpxchg(&sem->count, &count, count + adjust))
246 			break;
247 		if ((count & LDSEM_ACTIVE_MASK) == LDSEM_ACTIVE_BIAS) {
248 			raw_spin_unlock_irq(&sem->wait_lock);
249 			return sem;
250 		}
251 	} while (1);
252 
253 	list_add_tail(&waiter.list, &sem->write_wait);
254 
255 	waiter.task = current;
256 
257 	set_current_state(TASK_UNINTERRUPTIBLE);
258 	for (;;) {
259 		if (!timeout)
260 			break;
261 		raw_spin_unlock_irq(&sem->wait_lock);
262 		timeout = schedule_timeout(timeout);
263 		raw_spin_lock_irq(&sem->wait_lock);
264 		set_current_state(TASK_UNINTERRUPTIBLE);
265 		locked = writer_trylock(sem);
266 		if (locked)
267 			break;
268 	}
269 
270 	if (!locked)
271 		atomic_long_add_return(-LDSEM_WAIT_BIAS, &sem->count);
272 	list_del(&waiter.list);
273 
274 	/*
275 	 * In case of timeout, wake up every reader who gave the right of way
276 	 * to writer. Prevent separation readers into two groups:
277 	 * one that helds semaphore and another that sleeps.
278 	 * (in case of no contention with a writer)
279 	 */
280 	if (!locked && list_empty(&sem->write_wait))
281 		__ldsem_wake_readers(sem);
282 
283 	raw_spin_unlock_irq(&sem->wait_lock);
284 
285 	__set_current_state(TASK_RUNNING);
286 
287 	/* lock wait may have timed out */
288 	if (!locked)
289 		return NULL;
290 	return sem;
291 }
292 
293 
294 
__ldsem_down_read_nested(struct ld_semaphore * sem,int subclass,long timeout)295 static int __ldsem_down_read_nested(struct ld_semaphore *sem,
296 					   int subclass, long timeout)
297 {
298 	long count;
299 
300 	rwsem_acquire_read(&sem->dep_map, subclass, 0, _RET_IP_);
301 
302 	count = atomic_long_add_return(LDSEM_READ_BIAS, &sem->count);
303 	if (count <= 0) {
304 		lock_contended(&sem->dep_map, _RET_IP_);
305 		if (!down_read_failed(sem, count, timeout)) {
306 			rwsem_release(&sem->dep_map, _RET_IP_);
307 			return 0;
308 		}
309 	}
310 	lock_acquired(&sem->dep_map, _RET_IP_);
311 	return 1;
312 }
313 
__ldsem_down_write_nested(struct ld_semaphore * sem,int subclass,long timeout)314 static int __ldsem_down_write_nested(struct ld_semaphore *sem,
315 					    int subclass, long timeout)
316 {
317 	long count;
318 
319 	rwsem_acquire(&sem->dep_map, subclass, 0, _RET_IP_);
320 
321 	count = atomic_long_add_return(LDSEM_WRITE_BIAS, &sem->count);
322 	if ((count & LDSEM_ACTIVE_MASK) != LDSEM_ACTIVE_BIAS) {
323 		lock_contended(&sem->dep_map, _RET_IP_);
324 		if (!down_write_failed(sem, count, timeout)) {
325 			rwsem_release(&sem->dep_map, _RET_IP_);
326 			return 0;
327 		}
328 	}
329 	lock_acquired(&sem->dep_map, _RET_IP_);
330 	return 1;
331 }
332 
333 
334 /*
335  * lock for reading -- returns 1 if successful, 0 if timed out
336  */
ldsem_down_read(struct ld_semaphore * sem,long timeout)337 int __sched ldsem_down_read(struct ld_semaphore *sem, long timeout)
338 {
339 	might_sleep();
340 	return __ldsem_down_read_nested(sem, 0, timeout);
341 }
342 
343 /*
344  * trylock for reading -- returns 1 if successful, 0 if contention
345  */
ldsem_down_read_trylock(struct ld_semaphore * sem)346 int ldsem_down_read_trylock(struct ld_semaphore *sem)
347 {
348 	long count = atomic_long_read(&sem->count);
349 
350 	while (count >= 0) {
351 		if (atomic_long_try_cmpxchg(&sem->count, &count, count + LDSEM_READ_BIAS)) {
352 			rwsem_acquire_read(&sem->dep_map, 0, 1, _RET_IP_);
353 			lock_acquired(&sem->dep_map, _RET_IP_);
354 			return 1;
355 		}
356 	}
357 	return 0;
358 }
359 
360 /*
361  * lock for writing -- returns 1 if successful, 0 if timed out
362  */
ldsem_down_write(struct ld_semaphore * sem,long timeout)363 int __sched ldsem_down_write(struct ld_semaphore *sem, long timeout)
364 {
365 	might_sleep();
366 	return __ldsem_down_write_nested(sem, 0, timeout);
367 }
368 
369 /*
370  * trylock for writing -- returns 1 if successful, 0 if contention
371  */
ldsem_down_write_trylock(struct ld_semaphore * sem)372 int ldsem_down_write_trylock(struct ld_semaphore *sem)
373 {
374 	long count = atomic_long_read(&sem->count);
375 
376 	while ((count & LDSEM_ACTIVE_MASK) == 0) {
377 		if (atomic_long_try_cmpxchg(&sem->count, &count, count + LDSEM_WRITE_BIAS)) {
378 			rwsem_acquire(&sem->dep_map, 0, 1, _RET_IP_);
379 			lock_acquired(&sem->dep_map, _RET_IP_);
380 			return 1;
381 		}
382 	}
383 	return 0;
384 }
385 
386 /*
387  * release a read lock
388  */
ldsem_up_read(struct ld_semaphore * sem)389 void ldsem_up_read(struct ld_semaphore *sem)
390 {
391 	long count;
392 
393 	rwsem_release(&sem->dep_map, _RET_IP_);
394 
395 	count = atomic_long_add_return(-LDSEM_READ_BIAS, &sem->count);
396 	if (count < 0 && (count & LDSEM_ACTIVE_MASK) == 0)
397 		ldsem_wake(sem);
398 }
399 
400 /*
401  * release a write lock
402  */
ldsem_up_write(struct ld_semaphore * sem)403 void ldsem_up_write(struct ld_semaphore *sem)
404 {
405 	long count;
406 
407 	rwsem_release(&sem->dep_map, _RET_IP_);
408 
409 	count = atomic_long_add_return(-LDSEM_WRITE_BIAS, &sem->count);
410 	if (count < 0)
411 		ldsem_wake(sem);
412 }
413 
414 
415 #ifdef CONFIG_DEBUG_LOCK_ALLOC
416 
ldsem_down_read_nested(struct ld_semaphore * sem,int subclass,long timeout)417 int ldsem_down_read_nested(struct ld_semaphore *sem, int subclass, long timeout)
418 {
419 	might_sleep();
420 	return __ldsem_down_read_nested(sem, subclass, timeout);
421 }
422 
ldsem_down_write_nested(struct ld_semaphore * sem,int subclass,long timeout)423 int ldsem_down_write_nested(struct ld_semaphore *sem, int subclass,
424 			    long timeout)
425 {
426 	might_sleep();
427 	return __ldsem_down_write_nested(sem, subclass, timeout);
428 }
429 
430 #endif
431