projects/linux·blob uaccess-cleanups · refs

1786 lines · 47.5 KiB

1// SPDX-License-Identifier: GPL-2.0
2/* kernel/rwsem.c: R/W semaphores, public implementation
3 *
4 * Written by David Howells ([email protected]).
5 * Derived from asm-i386/semaphore.h
6 *
7 * Writer lock-stealing by Alex Shi <[email protected]>
8 * and Michel Lespinasse <[email protected]>
9 *
10 * Optimistic spinning by Tim Chen <[email protected]>
11 * and Davidlohr Bueso <[email protected]>. Based on mutexes.
12 *
13 * Rwsem count bit fields re-definition and rwsem rearchitecture by
14 * Waiman Long <[email protected]> and
15 * Peter Zijlstra <[email protected]>.
16 */
17
18#include <linux/types.h>
19#include <linux/kernel.h>
20#include <linux/sched.h>
21#include <linux/sched/rt.h>
22#include <linux/sched/task.h>
23#include <linux/sched/debug.h>
24#include <linux/sched/wake_q.h>
25#include <linux/sched/signal.h>
26#include <linux/sched/clock.h>
27#include <linux/export.h>
28#include <linux/rwsem.h>
29#include <linux/atomic.h>
30#include <linux/hung_task.h>
31#include <trace/events/lock.h>
32
33#ifndef CONFIG_PREEMPT_RT
34#include "lock_events.h"
35
36/*
37 * The least significant 2 bits of the owner value has the following
38 * meanings when set.
39 * - Bit 0: RWSEM_READER_OWNED - rwsem may be owned by readers (just a hint)
40 * - Bit 1: RWSEM_NONSPINNABLE - Cannot spin on a reader-owned lock
41 *
42 * When the rwsem is reader-owned and a spinning writer has timed out,
43 * the nonspinnable bit will be set to disable optimistic spinning.
44
45 * When a writer acquires a rwsem, it puts its task_struct pointer
46 * into the owner field. It is cleared after an unlock.
47 *
48 * When a reader acquires a rwsem, it will also puts its task_struct
49 * pointer into the owner field with the RWSEM_READER_OWNED bit set.
50 * On unlock, the owner field will largely be left untouched. So
51 * for a free or reader-owned rwsem, the owner value may contain
52 * information about the last reader that acquires the rwsem.
53 *
54 * That information may be helpful in debugging cases where the system
55 * seems to hang on a reader owned rwsem especially if only one reader
56 * is involved. Ideally we would like to track all the readers that own
57 * a rwsem, but the overhead is simply too big.
58 *
59 * A fast path reader optimistic lock stealing is supported when the rwsem
60 * is previously owned by a writer and the following conditions are met:
61 * - rwsem is not currently writer owned
62 * - the handoff isn't set.
63 */
64#define RWSEM_READER_OWNED (1UL << 0)
65#define RWSEM_NONSPINNABLE (1UL << 1)
66#define RWSEM_OWNER_FLAGS_MASK (RWSEM_READER_OWNED | RWSEM_NONSPINNABLE)
67
68#ifdef CONFIG_DEBUG_RWSEMS
69# define DEBUG_RWSEMS_WARN_ON(c, sem) do { \
70 if (!debug_locks_silent && \
71 WARN_ONCE(c, "DEBUG_RWSEMS_WARN_ON(%s): count = 0x%lx, magic = 0x%lx, owner = 0x%lx, curr 0x%lx, list %sempty\n",\
72 #c, atomic_long_read(&(sem)->count), \
73 (unsigned long) sem->magic, \
74 atomic_long_read(&(sem)->owner), (long)current, \
75 rwsem_is_contended(sem) ? "" : "not ")) \
76 debug_locks_off(); \
77 } while (0)
78#else
79# define DEBUG_RWSEMS_WARN_ON(c, sem)
80#endif
81
82/*
83 * On 64-bit architectures, the bit definitions of the count are:
84 *
85 * Bit 0 - writer locked bit
86 * Bit 1 - waiters present bit
87 * Bit 2 - lock handoff bit
88 * Bits 3-7 - reserved
89 * Bits 8-62 - 55-bit reader count
90 * Bit 63 - read fail bit
91 *
92 * On 32-bit architectures, the bit definitions of the count are:
93 *
94 * Bit 0 - writer locked bit
95 * Bit 1 - waiters present bit
96 * Bit 2 - lock handoff bit
97 * Bits 3-7 - reserved
98 * Bits 8-30 - 23-bit reader count
99 * Bit 31 - read fail bit
100 *
101 * It is not likely that the most significant bit (read fail bit) will ever
102 * be set. This guard bit is still checked anyway in the down_read() fastpath
103 * just in case we need to use up more of the reader bits for other purpose
104 * in the future.
105 *
106 * atomic_long_fetch_add() is used to obtain reader lock, whereas
107 * atomic_long_cmpxchg() will be used to obtain writer lock.
108 *
109 * There are three places where the lock handoff bit may be set or cleared.
110 * 1) rwsem_mark_wake() for readers -- set, clear
111 * 2) rwsem_try_write_lock() for writers -- set, clear
112 * 3) rwsem_del_waiter() -- clear
113 *
114 * For all the above cases, wait_lock will be held. A writer must also
115 * be the first one in the wait_list to be eligible for setting the handoff
116 * bit. So concurrent setting/clearing of handoff bit is not possible.
117 */
118#define RWSEM_WRITER_LOCKED (1UL << 0)
119#define RWSEM_FLAG_WAITERS (1UL << 1)
120#define RWSEM_FLAG_HANDOFF (1UL << 2)
121#define RWSEM_FLAG_READFAIL (1UL << (BITS_PER_LONG - 1))
122
123#define RWSEM_READER_SHIFT 8
124#define RWSEM_READER_BIAS (1UL << RWSEM_READER_SHIFT)
125#define RWSEM_READER_MASK (~(RWSEM_READER_BIAS - 1))
126#define RWSEM_WRITER_MASK RWSEM_WRITER_LOCKED
127#define RWSEM_LOCK_MASK (RWSEM_WRITER_MASK|RWSEM_READER_MASK)
128#define RWSEM_READ_FAILED_MASK (RWSEM_WRITER_MASK|RWSEM_FLAG_WAITERS|\
129 RWSEM_FLAG_HANDOFF|RWSEM_FLAG_READFAIL)
130
131/*
132 * All writes to owner are protected by WRITE_ONCE() to make sure that
133 * store tearing can't happen as optimistic spinners may read and use
134 * the owner value concurrently without lock. Read from owner, however,
135 * may not need READ_ONCE() as long as the pointer value is only used
136 * for comparison and isn't being dereferenced.
137 *
138 * Both rwsem_{set,clear}_owner() functions should be in the same
139 * preempt disable section as the atomic op that changes sem->count.
140 */
141static inline void rwsem_set_owner(struct rw_semaphore *sem)
142{
143 lockdep_assert_preemption_disabled();
144 atomic_long_set(&sem->owner, (long)current);
145}
146
147static inline void rwsem_clear_owner(struct rw_semaphore *sem)
148{
149 lockdep_assert_preemption_disabled();
150 atomic_long_set(&sem->owner, 0);
151}
152
153/*
154 * Test the flags in the owner field.
155 */
156static inline bool rwsem_test_oflags(struct rw_semaphore *sem, long flags)
157{
158 return atomic_long_read(&sem->owner) & flags;
159}
160
161/*
162 * The task_struct pointer of the last owning reader will be left in
163 * the owner field.
164 *
165 * Note that the owner value just indicates the task has owned the rwsem
166 * previously, it may not be the real owner or one of the real owners
167 * anymore when that field is examined, so take it with a grain of salt.
168 *
169 * The reader non-spinnable bit is preserved.
170 */
171static inline void __rwsem_set_reader_owned(struct rw_semaphore *sem,
172 struct task_struct *owner)
173{
174 unsigned long val = (unsigned long)owner | RWSEM_READER_OWNED |
175 (atomic_long_read(&sem->owner) & RWSEM_NONSPINNABLE);
176
177 atomic_long_set(&sem->owner, val);
178}
179
180static inline void rwsem_set_reader_owned(struct rw_semaphore *sem)
181{
182 __rwsem_set_reader_owned(sem, current);
183}
184
185#if defined(CONFIG_DEBUG_RWSEMS) || defined(CONFIG_DETECT_HUNG_TASK_BLOCKER)
186/*
187 * Return just the real task structure pointer of the owner
188 */
189struct task_struct *rwsem_owner(struct rw_semaphore *sem)
190{
191 return (struct task_struct *)
192 (atomic_long_read(&sem->owner) & ~RWSEM_OWNER_FLAGS_MASK);
193}
194
195/*
196 * Return true if the rwsem is owned by a reader.
197 */
198bool is_rwsem_reader_owned(struct rw_semaphore *sem)
199{
200 /*
201 * Check the count to see if it is write-locked.
202 */
203 long count = atomic_long_read(&sem->count);
204
205 if (count & RWSEM_WRITER_MASK)
206 return false;
207 return rwsem_test_oflags(sem, RWSEM_READER_OWNED);
208}
209
210/*
211 * With CONFIG_DEBUG_RWSEMS or CONFIG_DETECT_HUNG_TASK_BLOCKER configured,
212 * it will make sure that the owner field of a reader-owned rwsem either
213 * points to a real reader-owner(s) or gets cleared. The only exception is
214 * when the unlock is done by up_read_non_owner().
215 */
216static inline void rwsem_clear_reader_owned(struct rw_semaphore *sem)
217{
218 unsigned long val = atomic_long_read(&sem->owner);
219
220 while ((val & ~RWSEM_OWNER_FLAGS_MASK) == (unsigned long)current) {
221 if (atomic_long_try_cmpxchg(&sem->owner, &val,
222 val & RWSEM_OWNER_FLAGS_MASK))
223 return;
224 }
225}
226#else
227static inline void rwsem_clear_reader_owned(struct rw_semaphore *sem)
228{
229}
230#endif
231
232/*
233 * Set the RWSEM_NONSPINNABLE bits if the RWSEM_READER_OWNED flag
234 * remains set. Otherwise, the operation will be aborted.
235 */
236static inline void rwsem_set_nonspinnable(struct rw_semaphore *sem)
237{
238 unsigned long owner = atomic_long_read(&sem->owner);
239
240 do {
241 if (!(owner & RWSEM_READER_OWNED))
242 break;
243 if (owner & RWSEM_NONSPINNABLE)
244 break;
245 } while (!atomic_long_try_cmpxchg(&sem->owner, &owner,
246 owner | RWSEM_NONSPINNABLE));
247}
248
249static inline bool rwsem_read_trylock(struct rw_semaphore *sem, long *cntp)
250{
251 *cntp = atomic_long_add_return_acquire(RWSEM_READER_BIAS, &sem->count);
252
253 if (WARN_ON_ONCE(*cntp < 0))
254 rwsem_set_nonspinnable(sem);
255
256 if (!(*cntp & RWSEM_READ_FAILED_MASK)) {
257 rwsem_set_reader_owned(sem);
258 return true;
259 }
260
261 return false;
262}
263
264static inline bool rwsem_write_trylock(struct rw_semaphore *sem)
265{
266 long tmp = RWSEM_UNLOCKED_VALUE;
267
268 if (atomic_long_try_cmpxchg_acquire(&sem->count, &tmp, RWSEM_WRITER_LOCKED)) {
269 rwsem_set_owner(sem);
270 return true;
271 }
272
273 return false;
274}
275
276/*
277 * Return the real task structure pointer of the owner and the embedded
278 * flags in the owner. pflags must be non-NULL.
279 */
280static inline struct task_struct *
281rwsem_owner_flags(struct rw_semaphore *sem, unsigned long *pflags)
282{
283 unsigned long owner = atomic_long_read(&sem->owner);
284
285 *pflags = owner & RWSEM_OWNER_FLAGS_MASK;
286 return (struct task_struct *)(owner & ~RWSEM_OWNER_FLAGS_MASK);
287}
288
289/*
290 * Guide to the rw_semaphore's count field.
291 *
292 * When the RWSEM_WRITER_LOCKED bit in count is set, the lock is owned
293 * by a writer.
294 *
295 * The lock is owned by readers when
296 * (1) the RWSEM_WRITER_LOCKED isn't set in count,
297 * (2) some of the reader bits are set in count, and
298 * (3) the owner field has RWSEM_READ_OWNED bit set.
299 *
300 * Having some reader bits set is not enough to guarantee a readers owned
301 * lock as the readers may be in the process of backing out from the count
302 * and a writer has just released the lock. So another writer may steal
303 * the lock immediately after that.
304 */
305
306/*
307 * Initialize an rwsem:
308 */
309void __init_rwsem(struct rw_semaphore *sem, const char *name,
310 struct lock_class_key *key)
311{
312#ifdef CONFIG_DEBUG_LOCK_ALLOC
313 /*
314 * Make sure we are not reinitializing a held semaphore:
315 */
316 debug_check_no_locks_freed((void *)sem, sizeof(*sem));
317 lockdep_init_map_wait(&sem->dep_map, name, key, 0, LD_WAIT_SLEEP);
318#endif
319#ifdef CONFIG_DEBUG_RWSEMS
320 sem->magic = sem;
321#endif
322 atomic_long_set(&sem->count, RWSEM_UNLOCKED_VALUE);
323 atomic_long_set(&sem->owner, 0L);
324 scoped_guard (raw_spinlock_init, &sem->wait_lock) {
325 sem->first_waiter = NULL;
326 }
327#ifdef CONFIG_RWSEM_SPIN_ON_OWNER
328 osq_lock_init(&sem->osq);
329#endif
330}
331EXPORT_SYMBOL(__init_rwsem);
332
333enum rwsem_waiter_type {
334 RWSEM_WAITING_FOR_WRITE,
335 RWSEM_WAITING_FOR_READ
336};
337
338struct rwsem_waiter {
339 struct list_head list;
340 struct task_struct *task;
341 enum rwsem_waiter_type type;
342 unsigned long timeout;
343 bool handoff_set;
344};
345
346enum rwsem_wake_type {
347 RWSEM_WAKE_ANY, /* Wake whatever's at head of wait list */
348 RWSEM_WAKE_READERS, /* Wake readers only */
349 RWSEM_WAKE_READ_OWNED /* Waker thread holds the read lock */
350};
351
352/*
353 * The typical HZ value is either 250 or 1000. So set the minimum waiting
354 * time to at least 4ms or 1 jiffy (if it is higher than 4ms) in the wait
355 * queue before initiating the handoff protocol.
356 */
357#define RWSEM_WAIT_TIMEOUT DIV_ROUND_UP(HZ, 250)
358
359/*
360 * Magic number to batch-wakeup waiting readers, even when writers are
361 * also present in the queue. This both limits the amount of work the
362 * waking thread must do and also prevents any potential counter overflow,
363 * however unlikely.
364 */
365#define MAX_READERS_WAKEUP 0x100
366
367static inline
368bool __rwsem_del_waiter(struct rw_semaphore *sem, struct rwsem_waiter *waiter)
369 __must_hold(&sem->wait_lock)
370{
371 if (list_empty(&waiter->list)) {
372 sem->first_waiter = NULL;
373 return false;
374 }
375
376 if (sem->first_waiter == waiter) {
377 sem->first_waiter = list_first_entry(&waiter->list,
378 struct rwsem_waiter, list);
379 }
380 list_del(&waiter->list);
381
382 return true;
383}
384
385/*
386 * Remove a waiter from the wait_list and clear flags.
387 *
388 * Both rwsem_mark_wake() and rwsem_try_write_lock() contain a full 'copy' of
389 * this function. Modify with care.
390 *
391 * Return: true if wait_list isn't empty and false otherwise
392 */
393static inline bool
394rwsem_del_waiter(struct rw_semaphore *sem, struct rwsem_waiter *waiter)
395{
396 lockdep_assert_held(&sem->wait_lock);
397 if (__rwsem_del_waiter(sem, waiter))
398 return true;
399 atomic_long_andnot(RWSEM_FLAG_HANDOFF | RWSEM_FLAG_WAITERS, &sem->count);
400 return false;
401}
402
403static inline
404struct rwsem_waiter *next_waiter(const struct rw_semaphore *sem,
405 const struct rwsem_waiter *waiter)
406 __must_hold(&sem->wait_lock)
407{
408 struct rwsem_waiter *next = list_first_entry(&waiter->list,
409 struct rwsem_waiter, list);
410 if (next == sem->first_waiter)
411 return NULL;
412 return next;
413}
414
415/*
416 * handle the lock release when processes blocked on it that can now run
417 * - if we come here from up_xxxx(), then the RWSEM_FLAG_WAITERS bit must
418 * have been set.
419 * - there must be someone on the queue
420 * - the wait_lock must be held by the caller
421 * - tasks are marked for wakeup, the caller must later invoke wake_up_q()
422 * to actually wakeup the blocked task(s) and drop the reference count,
423 * preferably when the wait_lock is released
424 * - woken process blocks are discarded from the list after having task zeroed
425 * - writers are only marked woken if downgrading is false
426 *
427 * Implies rwsem_del_waiter() for all woken readers.
428 */
429static void rwsem_mark_wake(struct rw_semaphore *sem,
430 enum rwsem_wake_type wake_type,
431 struct wake_q_head *wake_q)
432{
433 struct rwsem_waiter *waiter, *next;
434 long oldcount, woken = 0, adjustment = 0;
435 struct list_head wlist;
436
437 lockdep_assert_held(&sem->wait_lock);
438
439 /*
440 * Take a peek at the queue head waiter such that we can determine
441 * the wakeup(s) to perform.
442 */
443 waiter = sem->first_waiter;
444
445 if (waiter->type == RWSEM_WAITING_FOR_WRITE) {
446 if (wake_type == RWSEM_WAKE_ANY) {
447 /*
448 * Mark writer at the front of the queue for wakeup.
449 * Until the task is actually later awoken later by
450 * the caller, other writers are able to steal it.
451 * Readers, on the other hand, will block as they
452 * will notice the queued writer.
453 */
454 wake_q_add(wake_q, waiter->task);
455 lockevent_inc(rwsem_wake_writer);
456 }
457
458 return;
459 }
460
461 /*
462 * No reader wakeup if there are too many of them already.
463 */
464 if (unlikely(atomic_long_read(&sem->count) < 0))
465 return;
466
467 /*
468 * Writers might steal the lock before we grant it to the next reader.
469 * We prefer to do the first reader grant before counting readers
470 * so we can bail out early if a writer stole the lock.
471 */
472 if (wake_type != RWSEM_WAKE_READ_OWNED) {
473 struct task_struct *owner;
474
475 adjustment = RWSEM_READER_BIAS;
476 oldcount = atomic_long_fetch_add(adjustment, &sem->count);
477 if (unlikely(oldcount & RWSEM_WRITER_MASK)) {
478 /*
479 * When we've been waiting "too" long (for writers
480 * to give up the lock), request a HANDOFF to
481 * force the issue.
482 */
483 if (time_after(jiffies, waiter->timeout)) {
484 if (!(oldcount & RWSEM_FLAG_HANDOFF)) {
485 adjustment -= RWSEM_FLAG_HANDOFF;
486 lockevent_inc(rwsem_rlock_handoff);
487 }
488 waiter->handoff_set = true;
489 }
490
491 atomic_long_add(-adjustment, &sem->count);
492 return;
493 }
494 /*
495 * Set it to reader-owned to give spinners an early
496 * indication that readers now have the lock.
497 * The reader nonspinnable bit seen at slowpath entry of
498 * the reader is copied over.
499 */
500 owner = waiter->task;
501 __rwsem_set_reader_owned(sem, owner);
502 }
503
504 /*
505 * Grant up to MAX_READERS_WAKEUP read locks to all the readers in the
506 * queue. We know that the woken will be at least 1 as we accounted
507 * for above. Note we increment the 'active part' of the count by the
508 * number of readers before waking any processes up.
509 *
510 * This is an adaptation of the phase-fair R/W locks where at the
511 * reader phase (first waiter is a reader), all readers are eligible
512 * to acquire the lock at the same time irrespective of their order
513 * in the queue. The writers acquire the lock according to their
514 * order in the queue.
515 *
516 * We have to do wakeup in 2 passes to prevent the possibility that
517 * the reader count may be decremented before it is incremented. It
518 * is because the to-be-woken waiter may not have slept yet. So it
519 * may see waiter->task got cleared, finish its critical section and
520 * do an unlock before the reader count increment.
521 *
522 * 1) Collect the read-waiters in a separate list, count them and
523 * fully increment the reader count in rwsem.
524 * 2) For each waiters in the new list, clear waiter->task and
525 * put them into wake_q to be woken up later.
526 */
527 INIT_LIST_HEAD(&wlist);
528 do {
529 next = next_waiter(sem, waiter);
530 if (waiter->type == RWSEM_WAITING_FOR_WRITE)
531 continue;
532
533 woken++;
534 list_move_tail(&waiter->list, &wlist);
535 if (sem->first_waiter == waiter)
536 sem->first_waiter = next;
537
538 /*
539 * Limit # of readers that can be woken up per wakeup call.
540 */
541 if (unlikely(woken >= MAX_READERS_WAKEUP))
542 break;
543 } while ((waiter = next) != NULL);
544
545 adjustment = woken * RWSEM_READER_BIAS - adjustment;
546 lockevent_cond_inc(rwsem_wake_reader, woken);
547
548 oldcount = atomic_long_read(&sem->count);
549 if (!sem->first_waiter) {
550 /*
551 * Combined with list_move_tail() above, this implies
552 * rwsem_del_waiter().
553 */
554 adjustment -= RWSEM_FLAG_WAITERS;
555 if (oldcount & RWSEM_FLAG_HANDOFF)
556 adjustment -= RWSEM_FLAG_HANDOFF;
557 } else if (woken) {
558 /*
559 * When we've woken a reader, we no longer need to force
560 * writers to give up the lock and we can clear HANDOFF.
561 */
562 if (oldcount & RWSEM_FLAG_HANDOFF)
563 adjustment -= RWSEM_FLAG_HANDOFF;
564 }
565
566 if (adjustment)
567 atomic_long_add(adjustment, &sem->count);
568
569 /* 2nd pass */
570 list_for_each_entry_safe(waiter, next, &wlist, list) {
571 struct task_struct *tsk;
572
573 tsk = waiter->task;
574 get_task_struct(tsk);
575
576 /*
577 * Ensure calling get_task_struct() before setting the reader
578 * waiter to nil such that rwsem_down_read_slowpath() cannot
579 * race with do_exit() by always holding a reference count
580 * to the task to wakeup.
581 */
582 smp_store_release(&waiter->task, NULL);
583 /*
584 * Ensure issuing the wakeup (either by us or someone else)
585 * after setting the reader waiter to nil.
586 */
587 wake_q_add_safe(wake_q, tsk);
588 }
589}
590
591/*
592 * Remove a waiter and try to wake up other waiters in the wait queue
593 * This function is called from the out_nolock path of both the reader and
594 * writer slowpaths with wait_lock held. It releases the wait_lock and
595 * optionally wake up waiters before it returns.
596 */
597static inline void
598rwsem_del_wake_waiter(struct rw_semaphore *sem, struct rwsem_waiter *waiter,
599 struct wake_q_head *wake_q)
600 __releases(&sem->wait_lock)
601{
602 bool first = sem->first_waiter == waiter;
603
604 wake_q_init(wake_q);
605
606 /*
607 * If the wait_list isn't empty and the waiter to be deleted is
608 * the first waiter, we wake up the remaining waiters as they may
609 * be eligible to acquire or spin on the lock.
610 */
611 if (rwsem_del_waiter(sem, waiter) && first)
612 rwsem_mark_wake(sem, RWSEM_WAKE_ANY, wake_q);
613 raw_spin_unlock_irq(&sem->wait_lock);
614 if (!wake_q_empty(wake_q))
615 wake_up_q(wake_q);
616}
617
618/*
619 * This function must be called with the sem->wait_lock held to prevent
620 * race conditions between checking the rwsem wait list and setting the
621 * sem->count accordingly.
622 *
623 * Implies rwsem_del_waiter() on success.
624 */
625static inline bool rwsem_try_write_lock(struct rw_semaphore *sem,
626 struct rwsem_waiter *waiter)
627 __must_hold(&sem->wait_lock)
628{
629 struct rwsem_waiter *first = sem->first_waiter;
630 long count, new;
631
632 lockdep_assert_held(&sem->wait_lock);
633
634 count = atomic_long_read(&sem->count);
635 do {
636 bool has_handoff = !!(count & RWSEM_FLAG_HANDOFF);
637
638 if (has_handoff) {
639 /*
640 * Honor handoff bit and yield only when the first
641 * waiter is the one that set it. Otherwisee, we
642 * still try to acquire the rwsem.
643 */
644 if (first->handoff_set && (waiter != first))
645 return false;
646 }
647
648 new = count;
649
650 if (count & RWSEM_LOCK_MASK) {
651 /*
652 * A waiter (first or not) can set the handoff bit
653 * if it is an RT task or wait in the wait queue
654 * for too long.
655 */
656 if (has_handoff || (!rt_or_dl_task(waiter->task) &&
657 !time_after(jiffies, waiter->timeout)))
658 return false;
659
660 new |= RWSEM_FLAG_HANDOFF;
661 } else {
662 new |= RWSEM_WRITER_LOCKED;
663 new &= ~RWSEM_FLAG_HANDOFF;
664
665 if (list_empty(&first->list))
666 new &= ~RWSEM_FLAG_WAITERS;
667 }
668 } while (!atomic_long_try_cmpxchg_acquire(&sem->count, &count, new));
669
670 /*
671 * We have either acquired the lock with handoff bit cleared or set
672 * the handoff bit. Only the first waiter can have its handoff_set
673 * set here to enable optimistic spinning in slowpath loop.
674 */
675 if (new & RWSEM_FLAG_HANDOFF) {
676 first->handoff_set = true;
677 lockevent_inc(rwsem_wlock_handoff);
678 return false;
679 }
680
681 /*
682 * Have rwsem_try_write_lock() fully imply rwsem_del_waiter() on
683 * success.
684 */
685 __rwsem_del_waiter(sem, waiter);
686
687 rwsem_set_owner(sem);
688 return true;
689}
690
691/*
692 * The rwsem_spin_on_owner() function returns the following 4 values
693 * depending on the lock owner state.
694 * OWNER_NULL : owner is currently NULL
695 * OWNER_WRITER: when owner changes and is a writer
696 * OWNER_READER: when owner changes and the new owner may be a reader.
697 * OWNER_NONSPINNABLE:
698 * when optimistic spinning has to stop because either the
699 * owner stops running, is unknown, or its timeslice has
700 * been used up.
701 */
702enum owner_state {
703 OWNER_NULL = 1 << 0,
704 OWNER_WRITER = 1 << 1,
705 OWNER_READER = 1 << 2,
706 OWNER_NONSPINNABLE = 1 << 3,
707};
708
709#ifdef CONFIG_RWSEM_SPIN_ON_OWNER
710/*
711 * Try to acquire write lock before the writer has been put on wait queue.
712 */
713static inline bool rwsem_try_write_lock_unqueued(struct rw_semaphore *sem)
714{
715 long count = atomic_long_read(&sem->count);
716
717 while (!(count & (RWSEM_LOCK_MASK|RWSEM_FLAG_HANDOFF))) {
718 if (atomic_long_try_cmpxchg_acquire(&sem->count, &count,
719 count | RWSEM_WRITER_LOCKED)) {
720 rwsem_set_owner(sem);
721 lockevent_inc(rwsem_opt_lock);
722 return true;
723 }
724 }
725 return false;
726}
727
728static inline bool rwsem_can_spin_on_owner(struct rw_semaphore *sem)
729{
730 struct task_struct *owner;
731 unsigned long flags;
732 bool ret = true;
733
734 if (need_resched()) {
735 lockevent_inc(rwsem_opt_fail);
736 return false;
737 }
738
739 /*
740 * Disable preemption is equal to the RCU read-side crital section,
741 * thus the task_strcut structure won't go away.
742 */
743 owner = rwsem_owner_flags(sem, &flags);
744 /*
745 * Don't check the read-owner as the entry may be stale.
746 */
747 if ((flags & RWSEM_NONSPINNABLE) ||
748 (owner && !(flags & RWSEM_READER_OWNED) && !owner_on_cpu(owner)))
749 ret = false;
750
751 lockevent_cond_inc(rwsem_opt_fail, !ret);
752 return ret;
753}
754
755static inline enum owner_state
756rwsem_owner_state(struct task_struct *owner, unsigned long flags)
757{
758 if (flags & RWSEM_NONSPINNABLE)
759 return OWNER_NONSPINNABLE;
760
761 if (flags & RWSEM_READER_OWNED)
762 return OWNER_READER;
763
764 return owner ? OWNER_WRITER : OWNER_NULL;
765}
766
767static noinline enum owner_state
768rwsem_spin_on_owner(struct rw_semaphore *sem)
769{
770 struct task_struct *new, *owner;
771 unsigned long flags, new_flags;
772 enum owner_state state;
773
774 lockdep_assert_preemption_disabled();
775
776 owner = rwsem_owner_flags(sem, &flags);
777 state = rwsem_owner_state(owner, flags);
778 if (state != OWNER_WRITER)
779 return state;
780
781 for (;;) {
782 /*
783 * When a waiting writer set the handoff flag, it may spin
784 * on the owner as well. Once that writer acquires the lock,
785 * we can spin on it. So we don't need to quit even when the
786 * handoff bit is set.
787 */
788 new = rwsem_owner_flags(sem, &new_flags);
789 if ((new != owner) || (new_flags != flags)) {
790 state = rwsem_owner_state(new, new_flags);
791 break;
792 }
793
794 /*
795 * Ensure we emit the owner->on_cpu, dereference _after_
796 * checking sem->owner still matches owner, if that fails,
797 * owner might point to free()d memory, if it still matches,
798 * our spinning context already disabled preemption which is
799 * equal to RCU read-side crital section ensures the memory
800 * stays valid.
801 */
802 barrier();
803
804 if (need_resched() || !owner_on_cpu(owner)) {
805 state = OWNER_NONSPINNABLE;
806 break;
807 }
808
809 cpu_relax();
810 }
811
812 return state;
813}
814
815/*
816 * Calculate reader-owned rwsem spinning threshold for writer
817 *
818 * The more readers own the rwsem, the longer it will take for them to
819 * wind down and free the rwsem. So the empirical formula used to
820 * determine the actual spinning time limit here is:
821 *
822 * Spinning threshold = (10 + nr_readers/2)us
823 *
824 * The limit is capped to a maximum of 25us (30 readers). This is just
825 * a heuristic and is subjected to change in the future.
826 */
827static inline u64 rwsem_rspin_threshold(struct rw_semaphore *sem)
828{
829 long count = atomic_long_read(&sem->count);
830 int readers = count >> RWSEM_READER_SHIFT;
831 u64 delta;
832
833 if (readers > 30)
834 readers = 30;
835 delta = (20 + readers) * NSEC_PER_USEC / 2;
836
837 return sched_clock() + delta;
838}
839
840static bool rwsem_optimistic_spin(struct rw_semaphore *sem)
841{
842 bool taken = false;
843 int prev_owner_state = OWNER_NULL;
844 int loop = 0;
845 u64 rspin_threshold = 0;
846
847 /* sem->wait_lock should not be held when doing optimistic spinning */
848 if (!osq_lock(&sem->osq))
849 goto done;
850
851 /*
852 * Optimistically spin on the owner field and attempt to acquire the
853 * lock whenever the owner changes. Spinning will be stopped when:
854 * 1) the owning writer isn't running; or
855 * 2) readers own the lock and spinning time has exceeded limit.
856 */
857 for (;;) {
858 enum owner_state owner_state;
859
860 owner_state = rwsem_spin_on_owner(sem);
861 if (owner_state == OWNER_NONSPINNABLE)
862 break;
863
864 /*
865 * Try to acquire the lock
866 */
867 taken = rwsem_try_write_lock_unqueued(sem);
868
869 if (taken)
870 break;
871
872 /*
873 * Time-based reader-owned rwsem optimistic spinning
874 */
875 if (owner_state == OWNER_READER) {
876 /*
877 * Re-initialize rspin_threshold every time when
878 * the owner state changes from non-reader to reader.
879 * This allows a writer to steal the lock in between
880 * 2 reader phases and have the threshold reset at
881 * the beginning of the 2nd reader phase.
882 */
883 if (prev_owner_state != OWNER_READER) {
884 if (rwsem_test_oflags(sem, RWSEM_NONSPINNABLE))
885 break;
886 rspin_threshold = rwsem_rspin_threshold(sem);
887 loop = 0;
888 }
889
890 /*
891 * Check time threshold once every 16 iterations to
892 * avoid calling sched_clock() too frequently so
893 * as to reduce the average latency between the times
894 * when the lock becomes free and when the spinner
895 * is ready to do a trylock.
896 */
897 else if (!(++loop & 0xf) && (sched_clock() > rspin_threshold)) {
898 rwsem_set_nonspinnable(sem);
899 lockevent_inc(rwsem_opt_nospin);
900 break;
901 }
902 }
903
904 /*
905 * An RT task cannot do optimistic spinning if it cannot
906 * be sure the lock holder is running or live-lock may
907 * happen if the current task and the lock holder happen
908 * to run in the same CPU. However, aborting optimistic
909 * spinning while a NULL owner is detected may miss some
910 * opportunity where spinning can continue without causing
911 * problem.
912 *
913 * There are 2 possible cases where an RT task may be able
914 * to continue spinning.
915 *
916 * 1) The lock owner is in the process of releasing the
917 * lock, sem->owner is cleared but the lock has not
918 * been released yet.
919 * 2) The lock was free and owner cleared, but another
920 * task just comes in and acquire the lock before
921 * we try to get it. The new owner may be a spinnable
922 * writer.
923 *
924 * To take advantage of two scenarios listed above, the RT
925 * task is made to retry one more time to see if it can
926 * acquire the lock or continue spinning on the new owning
927 * writer. Of course, if the time lag is long enough or the
928 * new owner is not a writer or spinnable, the RT task will
929 * quit spinning.
930 *
931 * If the owner is a writer, the need_resched() check is
932 * done inside rwsem_spin_on_owner(). If the owner is not
933 * a writer, need_resched() check needs to be done here.
934 */
935 if (owner_state != OWNER_WRITER) {
936 if (need_resched())
937 break;
938 if (rt_or_dl_task(current) &&
939 (prev_owner_state != OWNER_WRITER))
940 break;
941 }
942 prev_owner_state = owner_state;
943
944 /*
945 * The cpu_relax() call is a compiler barrier which forces
946 * everything in this loop to be re-loaded. We don't need
947 * memory barriers as we'll eventually observe the right
948 * values at the cost of a few extra spins.
949 */
950 cpu_relax();
951 }
952 osq_unlock(&sem->osq);
953done:
954 lockevent_cond_inc(rwsem_opt_fail, !taken);
955 return taken;
956}
957
958/*
959 * Clear the owner's RWSEM_NONSPINNABLE bit if it is set. This should
960 * only be called when the reader count reaches 0.
961 */
962static inline void clear_nonspinnable(struct rw_semaphore *sem)
963{
964 if (unlikely(rwsem_test_oflags(sem, RWSEM_NONSPINNABLE)))
965 atomic_long_andnot(RWSEM_NONSPINNABLE, &sem->owner);
966}
967
968#else
969static inline bool rwsem_can_spin_on_owner(struct rw_semaphore *sem)
970{
971 return false;
972}
973
974static inline bool rwsem_optimistic_spin(struct rw_semaphore *sem)
975{
976 return false;
977}
978
979static inline void clear_nonspinnable(struct rw_semaphore *sem) { }
980
981static inline enum owner_state
982rwsem_spin_on_owner(struct rw_semaphore *sem)
983{
984 return OWNER_NONSPINNABLE;
985}
986#endif
987
988/*
989 * Prepare to wake up waiter(s) in the wait queue by putting them into the
990 * given wake_q if the rwsem lock owner isn't a writer. If rwsem is likely
991 * reader-owned, wake up read lock waiters in queue front or wake up any
992 * front waiter otherwise.
993
994 * This is being called from both reader and writer slow paths.
995 */
996static inline void rwsem_cond_wake_waiter(struct rw_semaphore *sem, long count,
997 struct wake_q_head *wake_q)
998{
999 enum rwsem_wake_type wake_type;
1000
1001 if (count & RWSEM_WRITER_MASK)
1002 return;
1003
1004 if (count & RWSEM_READER_MASK) {
1005 wake_type = RWSEM_WAKE_READERS;
1006 } else {
1007 wake_type = RWSEM_WAKE_ANY;
1008 clear_nonspinnable(sem);
1009 }
1010 rwsem_mark_wake(sem, wake_type, wake_q);
1011}
1012
1013/*
1014 * Wait for the read lock to be granted
1015 */
1016static struct rw_semaphore __sched *
1017rwsem_down_read_slowpath(struct rw_semaphore *sem, long count, unsigned int state)
1018{
1019 long adjustment = -RWSEM_READER_BIAS;
1020 long rcnt = (count >> RWSEM_READER_SHIFT);
1021 struct rwsem_waiter waiter, *first;
1022 DEFINE_WAKE_Q(wake_q);
1023
1024 /*
1025 * To prevent a constant stream of readers from starving a sleeping
1026 * writer, don't attempt optimistic lock stealing if the lock is
1027 * very likely owned by readers.
1028 */
1029 if ((atomic_long_read(&sem->owner) & RWSEM_READER_OWNED) &&
1030 (rcnt > 1) && !(count & RWSEM_WRITER_LOCKED))
1031 goto queue;
1032
1033 /*
1034 * Reader optimistic lock stealing.
1035 */
1036 if (!(count & (RWSEM_WRITER_LOCKED | RWSEM_FLAG_HANDOFF))) {
1037 rwsem_set_reader_owned(sem);
1038 lockevent_inc(rwsem_rlock_steal);
1039
1040 /*
1041 * Wake up other readers in the wait queue if it is
1042 * the first reader.
1043 */
1044 if ((rcnt == 1) && (count & RWSEM_FLAG_WAITERS)) {
1045 raw_spin_lock_irq(&sem->wait_lock);
1046 if (sem->first_waiter)
1047 rwsem_mark_wake(sem, RWSEM_WAKE_READ_OWNED,
1048 &wake_q);
1049 raw_spin_unlock_irq(&sem->wait_lock);
1050 wake_up_q(&wake_q);
1051 }
1052 return sem;
1053 }
1054
1055queue:
1056 waiter.task = current;
1057 waiter.type = RWSEM_WAITING_FOR_READ;
1058 waiter.timeout = jiffies + RWSEM_WAIT_TIMEOUT;
1059 waiter.handoff_set = false;
1060
1061 raw_spin_lock_irq(&sem->wait_lock);
1062 first = sem->first_waiter;
1063 if (!first) {
1064 /*
1065 * In case the wait queue is empty and the lock isn't owned
1066 * by a writer, this reader can exit the slowpath and return
1067 * immediately as its RWSEM_READER_BIAS has already been set
1068 * in the count.
1069 */
1070 if (!(atomic_long_read(&sem->count) & RWSEM_WRITER_MASK)) {
1071 /* Provide lock ACQUIRE */
1072 smp_acquire__after_ctrl_dep();
1073 raw_spin_unlock_irq(&sem->wait_lock);
1074 rwsem_set_reader_owned(sem);
1075 lockevent_inc(rwsem_rlock_fast);
1076 return sem;
1077 }
1078 adjustment += RWSEM_FLAG_WAITERS;
1079 INIT_LIST_HEAD(&waiter.list);
1080 sem->first_waiter = &waiter;
1081 } else {
1082 list_add_tail(&waiter.list, &first->list);
1083 }
1084
1085 /* we're now waiting on the lock, but no longer actively locking */
1086 count = atomic_long_add_return(adjustment, &sem->count);
1087
1088 rwsem_cond_wake_waiter(sem, count, &wake_q);
1089 raw_spin_unlock_irq(&sem->wait_lock);
1090
1091 if (!wake_q_empty(&wake_q))
1092 wake_up_q(&wake_q);
1093
1094 trace_contention_begin(sem, LCB_F_READ);
1095 set_current_state(state);
1096
1097 if (state == TASK_UNINTERRUPTIBLE)
1098 hung_task_set_blocker(sem, BLOCKER_TYPE_RWSEM_READER);
1099
1100 /* wait to be given the lock */
1101 for (;;) {
1102 if (!smp_load_acquire(&waiter.task)) {
1103 /* Matches rwsem_mark_wake()'s smp_store_release(). */
1104 break;
1105 }
1106 if (signal_pending_state(state, current)) {
1107 raw_spin_lock_irq(&sem->wait_lock);
1108 if (waiter.task)
1109 goto out_nolock;
1110 raw_spin_unlock_irq(&sem->wait_lock);
1111 /* Ordered by sem->wait_lock against rwsem_mark_wake(). */
1112 break;
1113 }
1114 schedule_preempt_disabled();
1115 lockevent_inc(rwsem_sleep_reader);
1116 set_current_state(state);
1117 }
1118
1119 if (state == TASK_UNINTERRUPTIBLE)
1120 hung_task_clear_blocker();
1121
1122 __set_current_state(TASK_RUNNING);
1123 lockevent_inc(rwsem_rlock);
1124 trace_contention_end(sem, 0);
1125 return sem;
1126
1127out_nolock:
1128 rwsem_del_wake_waiter(sem, &waiter, &wake_q);
1129 __set_current_state(TASK_RUNNING);
1130 lockevent_inc(rwsem_rlock_fail);
1131 trace_contention_end(sem, -EINTR);
1132 return ERR_PTR(-EINTR);
1133}
1134
1135/*
1136 * Wait until we successfully acquire the write lock
1137 */
1138static struct rw_semaphore __sched *
1139rwsem_down_write_slowpath(struct rw_semaphore *sem, int state)
1140{
1141 struct rwsem_waiter waiter, *first;
1142 DEFINE_WAKE_Q(wake_q);
1143
1144 /* do optimistic spinning and steal lock if possible */
1145 if (rwsem_can_spin_on_owner(sem) && rwsem_optimistic_spin(sem)) {
1146 /* rwsem_optimistic_spin() implies ACQUIRE on success */
1147 return sem;
1148 }
1149
1150 /*
1151 * Optimistic spinning failed, proceed to the slowpath
1152 * and block until we can acquire the sem.
1153 */
1154 waiter.task = current;
1155 waiter.type = RWSEM_WAITING_FOR_WRITE;
1156 waiter.timeout = jiffies + RWSEM_WAIT_TIMEOUT;
1157 waiter.handoff_set = false;
1158
1159 raw_spin_lock_irq(&sem->wait_lock);
1160
1161 first = sem->first_waiter;
1162 if (first) {
1163 list_add_tail(&waiter.list, &first->list);
1164 rwsem_cond_wake_waiter(sem, atomic_long_read(&sem->count),
1165 &wake_q);
1166 if (!wake_q_empty(&wake_q)) {
1167 /*
1168 * We want to minimize wait_lock hold time especially
1169 * when a large number of readers are to be woken up.
1170 */
1171 raw_spin_unlock_irq(&sem->wait_lock);
1172 wake_up_q(&wake_q);
1173 raw_spin_lock_irq(&sem->wait_lock);
1174 }
1175 } else {
1176 INIT_LIST_HEAD(&waiter.list);
1177 sem->first_waiter = &waiter;
1178 atomic_long_or(RWSEM_FLAG_WAITERS, &sem->count);
1179 }
1180
1181 /* wait until we successfully acquire the lock */
1182 set_current_state(state);
1183 trace_contention_begin(sem, LCB_F_WRITE);
1184
1185 if (state == TASK_UNINTERRUPTIBLE)
1186 hung_task_set_blocker(sem, BLOCKER_TYPE_RWSEM_WRITER);
1187
1188 for (;;) {
1189 if (rwsem_try_write_lock(sem, &waiter)) {
1190 /* rwsem_try_write_lock() implies ACQUIRE on success */
1191 break;
1192 }
1193
1194 raw_spin_unlock_irq(&sem->wait_lock);
1195
1196 if (signal_pending_state(state, current))
1197 goto out_nolock;
1198
1199 /*
1200 * After setting the handoff bit and failing to acquire
1201 * the lock, attempt to spin on owner to accelerate lock
1202 * transfer. If the previous owner is a on-cpu writer and it
1203 * has just released the lock, OWNER_NULL will be returned.
1204 * In this case, we attempt to acquire the lock again
1205 * without sleeping.
1206 */
1207 if (waiter.handoff_set) {
1208 enum owner_state owner_state;
1209
1210 owner_state = rwsem_spin_on_owner(sem);
1211 if (owner_state == OWNER_NULL)
1212 goto trylock_again;
1213 }
1214
1215 schedule_preempt_disabled();
1216 lockevent_inc(rwsem_sleep_writer);
1217 set_current_state(state);
1218trylock_again:
1219 raw_spin_lock_irq(&sem->wait_lock);
1220 }
1221
1222 if (state == TASK_UNINTERRUPTIBLE)
1223 hung_task_clear_blocker();
1224
1225 __set_current_state(TASK_RUNNING);
1226 raw_spin_unlock_irq(&sem->wait_lock);
1227 lockevent_inc(rwsem_wlock);
1228 trace_contention_end(sem, 0);
1229 return sem;
1230
1231out_nolock:
1232 __set_current_state(TASK_RUNNING);
1233 raw_spin_lock_irq(&sem->wait_lock);
1234 rwsem_del_wake_waiter(sem, &waiter, &wake_q);
1235 lockevent_inc(rwsem_wlock_fail);
1236 trace_contention_end(sem, -EINTR);
1237 return ERR_PTR(-EINTR);
1238}
1239
1240/*
1241 * handle waking up a waiter on the semaphore
1242 * - up_read/up_write has decremented the active part of count if we come here
1243 */
1244static struct rw_semaphore *rwsem_wake(struct rw_semaphore *sem)
1245{
1246 unsigned long flags;
1247 DEFINE_WAKE_Q(wake_q);
1248
1249 raw_spin_lock_irqsave(&sem->wait_lock, flags);
1250
1251 if (sem->first_waiter)
1252 rwsem_mark_wake(sem, RWSEM_WAKE_ANY, &wake_q);
1253
1254 raw_spin_unlock_irqrestore(&sem->wait_lock, flags);
1255 wake_up_q(&wake_q);
1256
1257 return sem;
1258}
1259
1260/*
1261 * downgrade a write lock into a read lock
1262 * - caller incremented waiting part of count and discovered it still negative
1263 * - just wake up any readers at the front of the queue
1264 */
1265static struct rw_semaphore *rwsem_downgrade_wake(struct rw_semaphore *sem)
1266{
1267 unsigned long flags;
1268 DEFINE_WAKE_Q(wake_q);
1269
1270 raw_spin_lock_irqsave(&sem->wait_lock, flags);
1271
1272 if (sem->first_waiter)
1273 rwsem_mark_wake(sem, RWSEM_WAKE_READ_OWNED, &wake_q);
1274
1275 raw_spin_unlock_irqrestore(&sem->wait_lock, flags);
1276 wake_up_q(&wake_q);
1277
1278 return sem;
1279}
1280
1281/*
1282 * lock for reading
1283 */
1284static __always_inline int __down_read_common(struct rw_semaphore *sem, int state)
1285{
1286 int ret = 0;
1287 long count;
1288
1289 preempt_disable();
1290 if (!rwsem_read_trylock(sem, &count)) {
1291 if (IS_ERR(rwsem_down_read_slowpath(sem, count, state))) {
1292 ret = -EINTR;
1293 goto out;
1294 }
1295 DEBUG_RWSEMS_WARN_ON(!is_rwsem_reader_owned(sem), sem);
1296 }
1297out:
1298 preempt_enable();
1299 return ret;
1300}
1301
1302static __always_inline void __down_read(struct rw_semaphore *sem)
1303{
1304 __down_read_common(sem, TASK_UNINTERRUPTIBLE);
1305}
1306
1307static __always_inline int __down_read_interruptible(struct rw_semaphore *sem)
1308{
1309 return __down_read_common(sem, TASK_INTERRUPTIBLE);
1310}
1311
1312static __always_inline int __down_read_killable(struct rw_semaphore *sem)
1313{
1314 return __down_read_common(sem, TASK_KILLABLE);
1315}
1316
1317static inline int __down_read_trylock(struct rw_semaphore *sem)
1318{
1319 int ret = 0;
1320 long tmp;
1321
1322 DEBUG_RWSEMS_WARN_ON(sem->magic != sem, sem);
1323
1324 preempt_disable();
1325 tmp = atomic_long_read(&sem->count);
1326 while (!(tmp & RWSEM_READ_FAILED_MASK)) {
1327 if (atomic_long_try_cmpxchg_acquire(&sem->count, &tmp,
1328 tmp + RWSEM_READER_BIAS)) {
1329 rwsem_set_reader_owned(sem);
1330 ret = 1;
1331 break;
1332 }
1333 }
1334 preempt_enable();
1335 return ret;
1336}
1337
1338/*
1339 * lock for writing
1340 */
1341static __always_inline int __down_write_common(struct rw_semaphore *sem, int state)
1342{
1343 int ret = 0;
1344
1345 preempt_disable();
1346 if (unlikely(!rwsem_write_trylock(sem))) {
1347 if (IS_ERR(rwsem_down_write_slowpath(sem, state)))
1348 ret = -EINTR;
1349 }
1350 preempt_enable();
1351 return ret;
1352}
1353
1354static __always_inline void __down_write(struct rw_semaphore *sem)
1355{
1356 __down_write_common(sem, TASK_UNINTERRUPTIBLE);
1357}
1358
1359static __always_inline int __down_write_killable(struct rw_semaphore *sem)
1360{
1361 return __down_write_common(sem, TASK_KILLABLE);
1362}
1363
1364static inline int __down_write_trylock(struct rw_semaphore *sem)
1365{
1366 int ret;
1367
1368 preempt_disable();
1369 DEBUG_RWSEMS_WARN_ON(sem->magic != sem, sem);
1370 ret = rwsem_write_trylock(sem);
1371 preempt_enable();
1372
1373 return ret;
1374}
1375
1376/*
1377 * unlock after reading
1378 */
1379static inline void __up_read(struct rw_semaphore *sem)
1380{
1381 long tmp;
1382
1383 DEBUG_RWSEMS_WARN_ON(sem->magic != sem, sem);
1384 DEBUG_RWSEMS_WARN_ON(!is_rwsem_reader_owned(sem), sem);
1385
1386 preempt_disable();
1387 rwsem_clear_reader_owned(sem);
1388 tmp = atomic_long_add_return_release(-RWSEM_READER_BIAS, &sem->count);
1389 DEBUG_RWSEMS_WARN_ON(tmp < 0, sem);
1390 if (trace_contended_release_enabled() && (tmp & RWSEM_FLAG_WAITERS))
1391 trace_call__contended_release(sem);
1392 if (unlikely((tmp & (RWSEM_LOCK_MASK|RWSEM_FLAG_WAITERS)) ==
1393 RWSEM_FLAG_WAITERS)) {
1394 clear_nonspinnable(sem);
1395 rwsem_wake(sem);
1396 }
1397 preempt_enable();
1398}
1399
1400/*
1401 * unlock after writing
1402 */
1403static inline void __up_write(struct rw_semaphore *sem)
1404{
1405 long tmp;
1406
1407 DEBUG_RWSEMS_WARN_ON(sem->magic != sem, sem);
1408 /*
1409 * sem->owner may differ from current if the ownership is transferred
1410 * to an anonymous writer by setting the RWSEM_NONSPINNABLE bits.
1411 */
1412 DEBUG_RWSEMS_WARN_ON((rwsem_owner(sem) != current) &&
1413 !rwsem_test_oflags(sem, RWSEM_NONSPINNABLE), sem);
1414
1415 preempt_disable();
1416 rwsem_clear_owner(sem);
1417 tmp = atomic_long_fetch_add_release(-RWSEM_WRITER_LOCKED, &sem->count);
1418 if (unlikely(tmp & RWSEM_FLAG_WAITERS)) {
1419 trace_contended_release(sem);
1420 rwsem_wake(sem);
1421 }
1422 preempt_enable();
1423}
1424
1425/*
1426 * downgrade write lock to read lock
1427 */
1428static inline void __downgrade_write(struct rw_semaphore *sem)
1429{
1430 long tmp;
1431
1432 /*
1433 * When downgrading from exclusive to shared ownership,
1434 * anything inside the write-locked region cannot leak
1435 * into the read side. In contrast, anything in the
1436 * read-locked region is ok to be re-ordered into the
1437 * write side. As such, rely on RELEASE semantics.
1438 */
1439 DEBUG_RWSEMS_WARN_ON(rwsem_owner(sem) != current, sem);
1440 preempt_disable();
1441 tmp = atomic_long_fetch_add_release(
1442 -RWSEM_WRITER_LOCKED+RWSEM_READER_BIAS, &sem->count);
1443 rwsem_set_reader_owned(sem);
1444 if (tmp & RWSEM_FLAG_WAITERS) {
1445 trace_contended_release(sem);
1446 rwsem_downgrade_wake(sem);
1447 }
1448 preempt_enable();
1449}
1450
1451#else /* !CONFIG_PREEMPT_RT */
1452
1453#define RT_MUTEX_BUILD_MUTEX
1454#include "rtmutex.c"
1455
1456#define rwbase_set_and_save_current_state(state) \
1457 set_current_state(state)
1458
1459#define rwbase_restore_current_state() \
1460 __set_current_state(TASK_RUNNING)
1461
1462#define rwbase_rtmutex_lock_state(rtm, state) \
1463 __rt_mutex_lock(rtm, state)
1464
1465#define rwbase_rtmutex_slowlock_locked(rtm, state, wq) \
1466 __rt_mutex_slowlock_locked(rtm, NULL, state, wq)
1467
1468#define rwbase_rtmutex_unlock(rtm) \
1469 __rt_mutex_unlock(rtm)
1470
1471#define rwbase_rtmutex_trylock(rtm) \
1472 __rt_mutex_trylock(rtm)
1473
1474#define rwbase_signal_pending_state(state, current) \
1475 signal_pending_state(state, current)
1476
1477#define rwbase_pre_schedule() \
1478 rt_mutex_pre_schedule()
1479
1480#define rwbase_schedule() \
1481 rt_mutex_schedule()
1482
1483#define rwbase_post_schedule() \
1484 rt_mutex_post_schedule()
1485
1486#include "rwbase_rt.c"
1487
1488void __init_rwsem(struct rw_semaphore *sem, const char *name,
1489 struct lock_class_key *key)
1490{
1491 init_rwbase_rt(&(sem)->rwbase);
1492
1493#ifdef CONFIG_DEBUG_LOCK_ALLOC
1494 debug_check_no_locks_freed((void *)sem, sizeof(*sem));
1495 lockdep_init_map_wait(&sem->dep_map, name, key, 0, LD_WAIT_SLEEP);
1496#endif
1497}
1498EXPORT_SYMBOL(__init_rwsem);
1499
1500static inline void __down_read(struct rw_semaphore *sem)
1501{
1502 rwbase_read_lock(&sem->rwbase, TASK_UNINTERRUPTIBLE);
1503}
1504
1505static inline int __down_read_interruptible(struct rw_semaphore *sem)
1506{
1507 return rwbase_read_lock(&sem->rwbase, TASK_INTERRUPTIBLE);
1508}
1509
1510static inline int __down_read_killable(struct rw_semaphore *sem)
1511{
1512 return rwbase_read_lock(&sem->rwbase, TASK_KILLABLE);
1513}
1514
1515static inline int __down_read_trylock(struct rw_semaphore *sem)
1516{
1517 return rwbase_read_trylock(&sem->rwbase);
1518}
1519
1520static inline void __up_read(struct rw_semaphore *sem)
1521{
1522 rwbase_read_unlock(&sem->rwbase, TASK_NORMAL);
1523}
1524
1525static inline void __sched __down_write(struct rw_semaphore *sem)
1526{
1527 rwbase_write_lock(&sem->rwbase, TASK_UNINTERRUPTIBLE);
1528}
1529
1530static inline int __sched __down_write_killable(struct rw_semaphore *sem)
1531{
1532 return rwbase_write_lock(&sem->rwbase, TASK_KILLABLE);
1533}
1534
1535static inline int __down_write_trylock(struct rw_semaphore *sem)
1536{
1537 return rwbase_write_trylock(&sem->rwbase);
1538}
1539
1540static inline void __up_write(struct rw_semaphore *sem)
1541{
1542 rwbase_write_unlock(&sem->rwbase);
1543}
1544
1545static inline void __downgrade_write(struct rw_semaphore *sem)
1546{
1547 rwbase_write_downgrade(&sem->rwbase);
1548}
1549
1550/* Debug stubs for the common API */
1551#define DEBUG_RWSEMS_WARN_ON(c, sem)
1552
1553static inline void __rwsem_set_reader_owned(struct rw_semaphore *sem,
1554 struct task_struct *owner)
1555{
1556}
1557
1558static inline bool is_rwsem_reader_owned(struct rw_semaphore *sem)
1559{
1560 int count = atomic_read(&sem->rwbase.readers);
1561
1562 return count < 0 && count != READER_BIAS;
1563}
1564
1565#endif /* CONFIG_PREEMPT_RT */
1566
1567/*
1568 * lock for reading
1569 */
1570void __sched down_read(struct rw_semaphore *sem)
1571 __no_context_analysis
1572{
1573 might_sleep();
1574 rwsem_acquire_read(&sem->dep_map, 0, 0, _RET_IP_);
1575
1576 LOCK_CONTENDED(sem, __down_read_trylock, __down_read);
1577}
1578EXPORT_SYMBOL(down_read);
1579
1580int __sched down_read_interruptible(struct rw_semaphore *sem)
1581 __no_context_analysis
1582{
1583 might_sleep();
1584 rwsem_acquire_read(&sem->dep_map, 0, 0, _RET_IP_);
1585
1586 if (LOCK_CONTENDED_RETURN(sem, __down_read_trylock, __down_read_interruptible)) {
1587 rwsem_release(&sem->dep_map, _RET_IP_);
1588 return -EINTR;
1589 }
1590
1591 return 0;
1592}
1593EXPORT_SYMBOL(down_read_interruptible);
1594
1595int __sched down_read_killable(struct rw_semaphore *sem)
1596 __no_context_analysis
1597{
1598 might_sleep();
1599 rwsem_acquire_read(&sem->dep_map, 0, 0, _RET_IP_);
1600
1601 if (LOCK_CONTENDED_RETURN(sem, __down_read_trylock, __down_read_killable)) {
1602 rwsem_release(&sem->dep_map, _RET_IP_);
1603 return -EINTR;
1604 }
1605
1606 return 0;
1607}
1608EXPORT_SYMBOL(down_read_killable);
1609
1610/*
1611 * trylock for reading -- returns 1 if successful, 0 if contention
1612 */
1613int down_read_trylock(struct rw_semaphore *sem)
1614 __no_context_analysis
1615{
1616 int ret = __down_read_trylock(sem);
1617
1618 if (ret == 1)
1619 rwsem_acquire_read(&sem->dep_map, 0, 1, _RET_IP_);
1620 return ret;
1621}
1622EXPORT_SYMBOL(down_read_trylock);
1623
1624/*
1625 * lock for writing
1626 */
1627void __sched down_write(struct rw_semaphore *sem)
1628 __no_context_analysis
1629{
1630 might_sleep();
1631 rwsem_acquire(&sem->dep_map, 0, 0, _RET_IP_);
1632 LOCK_CONTENDED(sem, __down_write_trylock, __down_write);
1633}
1634EXPORT_SYMBOL(down_write);
1635
1636/*
1637 * lock for writing
1638 */
1639int __sched down_write_killable(struct rw_semaphore *sem)
1640 __no_context_analysis
1641{
1642 might_sleep();
1643 rwsem_acquire(&sem->dep_map, 0, 0, _RET_IP_);
1644
1645 if (LOCK_CONTENDED_RETURN(sem, __down_write_trylock,
1646 __down_write_killable)) {
1647 rwsem_release(&sem->dep_map, _RET_IP_);
1648 return -EINTR;
1649 }
1650
1651 return 0;
1652}
1653EXPORT_SYMBOL(down_write_killable);
1654
1655/*
1656 * trylock for writing -- returns 1 if successful, 0 if contention
1657 */
1658int down_write_trylock(struct rw_semaphore *sem)
1659 __no_context_analysis
1660{
1661 int ret = __down_write_trylock(sem);
1662
1663 if (ret == 1)
1664 rwsem_acquire(&sem->dep_map, 0, 1, _RET_IP_);
1665
1666 return ret;
1667}
1668EXPORT_SYMBOL(down_write_trylock);
1669
1670/*
1671 * release a read lock
1672 */
1673void up_read(struct rw_semaphore *sem)
1674 __no_context_analysis
1675{
1676 rwsem_release(&sem->dep_map, _RET_IP_);
1677 __up_read(sem);
1678}
1679EXPORT_SYMBOL(up_read);
1680
1681/*
1682 * release a write lock
1683 */
1684void up_write(struct rw_semaphore *sem)
1685 __no_context_analysis
1686{
1687 rwsem_release(&sem->dep_map, _RET_IP_);
1688 __up_write(sem);
1689}
1690EXPORT_SYMBOL(up_write);
1691
1692/*
1693 * downgrade write lock to read lock
1694 */
1695void downgrade_write(struct rw_semaphore *sem)
1696 __no_context_analysis
1697{
1698 lock_downgrade(&sem->dep_map, _RET_IP_);
1699 __downgrade_write(sem);
1700}
1701EXPORT_SYMBOL(downgrade_write);
1702
1703#ifdef CONFIG_DEBUG_LOCK_ALLOC
1704
1705void down_read_nested(struct rw_semaphore *sem, int subclass)
1706 __no_context_analysis
1707{
1708 might_sleep();
1709 rwsem_acquire_read(&sem->dep_map, subclass, 0, _RET_IP_);
1710 LOCK_CONTENDED(sem, __down_read_trylock, __down_read);
1711}
1712EXPORT_SYMBOL(down_read_nested);
1713
1714int down_read_killable_nested(struct rw_semaphore *sem, int subclass)
1715 __no_context_analysis
1716{
1717 might_sleep();
1718 rwsem_acquire_read(&sem->dep_map, subclass, 0, _RET_IP_);
1719
1720 if (LOCK_CONTENDED_RETURN(sem, __down_read_trylock, __down_read_killable)) {
1721 rwsem_release(&sem->dep_map, _RET_IP_);
1722 return -EINTR;
1723 }
1724
1725 return 0;
1726}
1727EXPORT_SYMBOL(down_read_killable_nested);
1728
1729void _down_write_nest_lock(struct rw_semaphore *sem, struct lockdep_map *nest)
1730 __no_context_analysis
1731{
1732 might_sleep();
1733 rwsem_acquire_nest(&sem->dep_map, 0, 0, nest, _RET_IP_);
1734 LOCK_CONTENDED(sem, __down_write_trylock, __down_write);
1735}
1736EXPORT_SYMBOL(_down_write_nest_lock);
1737
1738void down_read_non_owner(struct rw_semaphore *sem)
1739 __no_context_analysis
1740{
1741 might_sleep();
1742 __down_read(sem);
1743 /*
1744 * The owner value for a reader-owned lock is mostly for debugging
1745 * purpose only and is not critical to the correct functioning of
1746 * rwsem. So it is perfectly fine to set it in a preempt-enabled
1747 * context here.
1748 */
1749 __rwsem_set_reader_owned(sem, NULL);
1750}
1751EXPORT_SYMBOL(down_read_non_owner);
1752
1753void down_write_nested(struct rw_semaphore *sem, int subclass)
1754 __no_context_analysis
1755{
1756 might_sleep();
1757 rwsem_acquire(&sem->dep_map, subclass, 0, _RET_IP_);
1758 LOCK_CONTENDED(sem, __down_write_trylock, __down_write);
1759}
1760EXPORT_SYMBOL(down_write_nested);
1761
1762int __sched down_write_killable_nested(struct rw_semaphore *sem, int subclass)
1763 __no_context_analysis
1764{
1765 might_sleep();
1766 rwsem_acquire(&sem->dep_map, subclass, 0, _RET_IP_);
1767
1768 if (LOCK_CONTENDED_RETURN(sem, __down_write_trylock,
1769 __down_write_killable)) {
1770 rwsem_release(&sem->dep_map, _RET_IP_);
1771 return -EINTR;
1772 }
1773
1774 return 0;
1775}
1776EXPORT_SYMBOL(down_write_killable_nested);
1777
1778void up_read_non_owner(struct rw_semaphore *sem)
1779 __no_context_analysis
1780{
1781 DEBUG_RWSEMS_WARN_ON(!is_rwsem_reader_owned(sem), sem);
1782 __up_read(sem);
1783}
1784EXPORT_SYMBOL(up_read_non_owner);
1785
1786#endif