1 // SPDX-License-Identifier: GPL-2.0
2 /* kernel/rwsem.c: R/W semaphores, public implementation
4 * Written by David Howells (dhowells@redhat.com).
5 * Derived from asm-i386/semaphore.h
7 * Writer lock-stealing by Alex Shi <alex.shi@intel.com>
8 * and Michel Lespinasse <walken@google.com>
10 * Optimistic spinning by Tim Chen <tim.c.chen@intel.com>
11 * and Davidlohr Bueso <davidlohr@hp.com>. Based on mutexes.
13 * Rwsem count bit fields re-definition and rwsem rearchitecture by
14 * Waiman Long <longman@redhat.com> and
15 * Peter Zijlstra <peterz@infradead.org>.
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>
31 #ifndef CONFIG_PREEMPT_RT
32 #include "lock_events.h"
35 * The least significant 2 bits of the owner value has the following
37 * - Bit 0: RWSEM_READER_OWNED - The rwsem is owned by readers
38 * - Bit 1: RWSEM_NONSPINNABLE - Cannot spin on a reader-owned lock
40 * When the rwsem is reader-owned and a spinning writer has timed out,
41 * the nonspinnable bit will be set to disable optimistic spinning.
43 * When a writer acquires a rwsem, it puts its task_struct pointer
44 * into the owner field. It is cleared after an unlock.
46 * When a reader acquires a rwsem, it will also puts its task_struct
47 * pointer into the owner field with the RWSEM_READER_OWNED bit set.
48 * On unlock, the owner field will largely be left untouched. So
49 * for a free or reader-owned rwsem, the owner value may contain
50 * information about the last reader that acquires the rwsem.
52 * That information may be helpful in debugging cases where the system
53 * seems to hang on a reader owned rwsem especially if only one reader
54 * is involved. Ideally we would like to track all the readers that own
55 * a rwsem, but the overhead is simply too big.
57 * A fast path reader optimistic lock stealing is supported when the rwsem
58 * is previously owned by a writer and the following conditions are met:
60 * - rwsem is not currently writer owned
61 * - the handoff isn't set.
63 #define RWSEM_READER_OWNED (1UL << 0)
64 #define RWSEM_NONSPINNABLE (1UL << 1)
65 #define RWSEM_OWNER_FLAGS_MASK (RWSEM_READER_OWNED | RWSEM_NONSPINNABLE)
67 #ifdef CONFIG_DEBUG_RWSEMS
68 # define DEBUG_RWSEMS_WARN_ON(c, sem) do { \
69 if (!debug_locks_silent && \
70 WARN_ONCE(c, "DEBUG_RWSEMS_WARN_ON(%s): count = 0x%lx, magic = 0x%lx, owner = 0x%lx, curr 0x%lx, list %sempty\n",\
71 #c, atomic_long_read(&(sem)->count), \
72 (unsigned long) sem->magic, \
73 atomic_long_read(&(sem)->owner), (long)current, \
74 list_empty(&(sem)->wait_list) ? "" : "not ")) \
78 # define DEBUG_RWSEMS_WARN_ON(c, sem)
82 * On 64-bit architectures, the bit definitions of the count are:
84 * Bit 0 - writer locked bit
85 * Bit 1 - waiters present bit
86 * Bit 2 - lock handoff bit
88 * Bits 8-62 - 55-bit reader count
89 * Bit 63 - read fail bit
91 * On 32-bit architectures, the bit definitions of the count are:
93 * Bit 0 - writer locked bit
94 * Bit 1 - waiters present bit
95 * Bit 2 - lock handoff bit
97 * Bits 8-30 - 23-bit reader count
98 * Bit 31 - read fail bit
100 * It is not likely that the most significant bit (read fail bit) will ever
101 * be set. This guard bit is still checked anyway in the down_read() fastpath
102 * just in case we need to use up more of the reader bits for other purpose
105 * atomic_long_fetch_add() is used to obtain reader lock, whereas
106 * atomic_long_cmpxchg() will be used to obtain writer lock.
108 * There are three places where the lock handoff bit may be set or cleared.
109 * 1) rwsem_mark_wake() for readers -- set, clear
110 * 2) rwsem_try_write_lock() for writers -- set, clear
111 * 3) rwsem_del_waiter() -- clear
113 * For all the above cases, wait_lock will be held. A writer must also
114 * be the first one in the wait_list to be eligible for setting the handoff
115 * bit. So concurrent setting/clearing of handoff bit is not possible.
117 #define RWSEM_WRITER_LOCKED (1UL << 0)
118 #define RWSEM_FLAG_WAITERS (1UL << 1)
119 #define RWSEM_FLAG_HANDOFF (1UL << 2)
120 #define RWSEM_FLAG_READFAIL (1UL << (BITS_PER_LONG - 1))
122 #define RWSEM_READER_SHIFT 8
123 #define RWSEM_READER_BIAS (1UL << RWSEM_READER_SHIFT)
124 #define RWSEM_READER_MASK (~(RWSEM_READER_BIAS - 1))
125 #define RWSEM_WRITER_MASK RWSEM_WRITER_LOCKED
126 #define RWSEM_LOCK_MASK (RWSEM_WRITER_MASK|RWSEM_READER_MASK)
127 #define RWSEM_READ_FAILED_MASK (RWSEM_WRITER_MASK|RWSEM_FLAG_WAITERS|\
128 RWSEM_FLAG_HANDOFF|RWSEM_FLAG_READFAIL)
131 * All writes to owner are protected by WRITE_ONCE() to make sure that
132 * store tearing can't happen as optimistic spinners may read and use
133 * the owner value concurrently without lock. Read from owner, however,
134 * may not need READ_ONCE() as long as the pointer value is only used
135 * for comparison and isn't being dereferenced.
137 static inline void rwsem_set_owner(struct rw_semaphore *sem)
139 atomic_long_set(&sem->owner, (long)current);
142 static inline void rwsem_clear_owner(struct rw_semaphore *sem)
144 atomic_long_set(&sem->owner, 0);
148 * Test the flags in the owner field.
150 static inline bool rwsem_test_oflags(struct rw_semaphore *sem, long flags)
152 return atomic_long_read(&sem->owner) & flags;
156 * The task_struct pointer of the last owning reader will be left in
159 * Note that the owner value just indicates the task has owned the rwsem
160 * previously, it may not be the real owner or one of the real owners
161 * anymore when that field is examined, so take it with a grain of salt.
163 * The reader non-spinnable bit is preserved.
165 static inline void __rwsem_set_reader_owned(struct rw_semaphore *sem,
166 struct task_struct *owner)
168 unsigned long val = (unsigned long)owner | RWSEM_READER_OWNED |
169 (atomic_long_read(&sem->owner) & RWSEM_NONSPINNABLE);
171 atomic_long_set(&sem->owner, val);
174 static inline void rwsem_set_reader_owned(struct rw_semaphore *sem)
176 __rwsem_set_reader_owned(sem, current);
180 * Return true if the rwsem is owned by a reader.
182 static inline bool is_rwsem_reader_owned(struct rw_semaphore *sem)
184 #ifdef CONFIG_DEBUG_RWSEMS
186 * Check the count to see if it is write-locked.
188 long count = atomic_long_read(&sem->count);
190 if (count & RWSEM_WRITER_MASK)
193 return rwsem_test_oflags(sem, RWSEM_READER_OWNED);
196 #ifdef CONFIG_DEBUG_RWSEMS
198 * With CONFIG_DEBUG_RWSEMS configured, it will make sure that if there
199 * is a task pointer in owner of a reader-owned rwsem, it will be the
200 * real owner or one of the real owners. The only exception is when the
201 * unlock is done by up_read_non_owner().
203 static inline void rwsem_clear_reader_owned(struct rw_semaphore *sem)
205 unsigned long val = atomic_long_read(&sem->owner);
207 while ((val & ~RWSEM_OWNER_FLAGS_MASK) == (unsigned long)current) {
208 if (atomic_long_try_cmpxchg(&sem->owner, &val,
209 val & RWSEM_OWNER_FLAGS_MASK))
214 static inline void rwsem_clear_reader_owned(struct rw_semaphore *sem)
220 * Set the RWSEM_NONSPINNABLE bits if the RWSEM_READER_OWNED flag
221 * remains set. Otherwise, the operation will be aborted.
223 static inline void rwsem_set_nonspinnable(struct rw_semaphore *sem)
225 unsigned long owner = atomic_long_read(&sem->owner);
228 if (!(owner & RWSEM_READER_OWNED))
230 if (owner & RWSEM_NONSPINNABLE)
232 } while (!atomic_long_try_cmpxchg(&sem->owner, &owner,
233 owner | RWSEM_NONSPINNABLE));
236 static inline bool rwsem_read_trylock(struct rw_semaphore *sem, long *cntp)
238 *cntp = atomic_long_add_return_acquire(RWSEM_READER_BIAS, &sem->count);
240 if (WARN_ON_ONCE(*cntp < 0))
241 rwsem_set_nonspinnable(sem);
243 if (!(*cntp & RWSEM_READ_FAILED_MASK)) {
244 rwsem_set_reader_owned(sem);
251 static inline bool rwsem_write_trylock(struct rw_semaphore *sem)
253 long tmp = RWSEM_UNLOCKED_VALUE;
255 if (atomic_long_try_cmpxchg_acquire(&sem->count, &tmp, RWSEM_WRITER_LOCKED)) {
256 rwsem_set_owner(sem);
264 * Return just the real task structure pointer of the owner
266 static inline struct task_struct *rwsem_owner(struct rw_semaphore *sem)
268 return (struct task_struct *)
269 (atomic_long_read(&sem->owner) & ~RWSEM_OWNER_FLAGS_MASK);
273 * Return the real task structure pointer of the owner and the embedded
274 * flags in the owner. pflags must be non-NULL.
276 static inline struct task_struct *
277 rwsem_owner_flags(struct rw_semaphore *sem, unsigned long *pflags)
279 unsigned long owner = atomic_long_read(&sem->owner);
281 *pflags = owner & RWSEM_OWNER_FLAGS_MASK;
282 return (struct task_struct *)(owner & ~RWSEM_OWNER_FLAGS_MASK);
286 * Guide to the rw_semaphore's count field.
288 * When the RWSEM_WRITER_LOCKED bit in count is set, the lock is owned
291 * The lock is owned by readers when
292 * (1) the RWSEM_WRITER_LOCKED isn't set in count,
293 * (2) some of the reader bits are set in count, and
294 * (3) the owner field has RWSEM_READ_OWNED bit set.
296 * Having some reader bits set is not enough to guarantee a readers owned
297 * lock as the readers may be in the process of backing out from the count
298 * and a writer has just released the lock. So another writer may steal
299 * the lock immediately after that.
303 * Initialize an rwsem:
305 void __init_rwsem(struct rw_semaphore *sem, const char *name,
306 struct lock_class_key *key)
308 #ifdef CONFIG_DEBUG_LOCK_ALLOC
310 * Make sure we are not reinitializing a held semaphore:
312 debug_check_no_locks_freed((void *)sem, sizeof(*sem));
313 lockdep_init_map_wait(&sem->dep_map, name, key, 0, LD_WAIT_SLEEP);
315 #ifdef CONFIG_DEBUG_RWSEMS
318 atomic_long_set(&sem->count, RWSEM_UNLOCKED_VALUE);
319 raw_spin_lock_init(&sem->wait_lock);
320 INIT_LIST_HEAD(&sem->wait_list);
321 atomic_long_set(&sem->owner, 0L);
322 #ifdef CONFIG_RWSEM_SPIN_ON_OWNER
323 osq_lock_init(&sem->osq);
326 EXPORT_SYMBOL(__init_rwsem);
328 enum rwsem_waiter_type {
329 RWSEM_WAITING_FOR_WRITE,
330 RWSEM_WAITING_FOR_READ
333 struct rwsem_waiter {
334 struct list_head list;
335 struct task_struct *task;
336 enum rwsem_waiter_type type;
337 unsigned long timeout;
340 #define rwsem_first_waiter(sem) \
341 list_first_entry(&sem->wait_list, struct rwsem_waiter, list)
343 enum rwsem_wake_type {
344 RWSEM_WAKE_ANY, /* Wake whatever's at head of wait list */
345 RWSEM_WAKE_READERS, /* Wake readers only */
346 RWSEM_WAKE_READ_OWNED /* Waker thread holds the read lock */
350 * The typical HZ value is either 250 or 1000. So set the minimum waiting
351 * time to at least 4ms or 1 jiffy (if it is higher than 4ms) in the wait
352 * queue before initiating the handoff protocol.
354 #define RWSEM_WAIT_TIMEOUT DIV_ROUND_UP(HZ, 250)
357 * Magic number to batch-wakeup waiting readers, even when writers are
358 * also present in the queue. This both limits the amount of work the
359 * waking thread must do and also prevents any potential counter overflow,
362 #define MAX_READERS_WAKEUP 0x100
365 rwsem_add_waiter(struct rw_semaphore *sem, struct rwsem_waiter *waiter)
367 lockdep_assert_held(&sem->wait_lock);
368 list_add_tail(&waiter->list, &sem->wait_list);
369 /* caller will set RWSEM_FLAG_WAITERS */
373 * Remove a waiter from the wait_list and clear flags.
375 * Both rwsem_mark_wake() and rwsem_try_write_lock() contain a full 'copy' of
376 * this function. Modify with care.
379 rwsem_del_waiter(struct rw_semaphore *sem, struct rwsem_waiter *waiter)
381 lockdep_assert_held(&sem->wait_lock);
382 list_del(&waiter->list);
383 if (likely(!list_empty(&sem->wait_list)))
386 atomic_long_andnot(RWSEM_FLAG_HANDOFF | RWSEM_FLAG_WAITERS, &sem->count);
390 * handle the lock release when processes blocked on it that can now run
391 * - if we come here from up_xxxx(), then the RWSEM_FLAG_WAITERS bit must
393 * - there must be someone on the queue
394 * - the wait_lock must be held by the caller
395 * - tasks are marked for wakeup, the caller must later invoke wake_up_q()
396 * to actually wakeup the blocked task(s) and drop the reference count,
397 * preferably when the wait_lock is released
398 * - woken process blocks are discarded from the list after having task zeroed
399 * - writers are only marked woken if downgrading is false
401 * Implies rwsem_del_waiter() for all woken readers.
403 static void rwsem_mark_wake(struct rw_semaphore *sem,
404 enum rwsem_wake_type wake_type,
405 struct wake_q_head *wake_q)
407 struct rwsem_waiter *waiter, *tmp;
408 long oldcount, woken = 0, adjustment = 0;
409 struct list_head wlist;
411 lockdep_assert_held(&sem->wait_lock);
414 * Take a peek at the queue head waiter such that we can determine
415 * the wakeup(s) to perform.
417 waiter = rwsem_first_waiter(sem);
419 if (waiter->type == RWSEM_WAITING_FOR_WRITE) {
420 if (wake_type == RWSEM_WAKE_ANY) {
422 * Mark writer at the front of the queue for wakeup.
423 * Until the task is actually later awoken later by
424 * the caller, other writers are able to steal it.
425 * Readers, on the other hand, will block as they
426 * will notice the queued writer.
428 wake_q_add(wake_q, waiter->task);
429 lockevent_inc(rwsem_wake_writer);
436 * No reader wakeup if there are too many of them already.
438 if (unlikely(atomic_long_read(&sem->count) < 0))
442 * Writers might steal the lock before we grant it to the next reader.
443 * We prefer to do the first reader grant before counting readers
444 * so we can bail out early if a writer stole the lock.
446 if (wake_type != RWSEM_WAKE_READ_OWNED) {
447 struct task_struct *owner;
449 adjustment = RWSEM_READER_BIAS;
450 oldcount = atomic_long_fetch_add(adjustment, &sem->count);
451 if (unlikely(oldcount & RWSEM_WRITER_MASK)) {
453 * When we've been waiting "too" long (for writers
454 * to give up the lock), request a HANDOFF to
457 if (time_after(jiffies, waiter->timeout)) {
458 if (!(oldcount & RWSEM_FLAG_HANDOFF)) {
459 adjustment -= RWSEM_FLAG_HANDOFF;
460 lockevent_inc(rwsem_rlock_handoff);
462 waiter->handoff_set = true;
465 atomic_long_add(-adjustment, &sem->count);
469 * Set it to reader-owned to give spinners an early
470 * indication that readers now have the lock.
471 * The reader nonspinnable bit seen at slowpath entry of
472 * the reader is copied over.
474 owner = waiter->task;
475 __rwsem_set_reader_owned(sem, owner);
479 * Grant up to MAX_READERS_WAKEUP read locks to all the readers in the
480 * queue. We know that the woken will be at least 1 as we accounted
481 * for above. Note we increment the 'active part' of the count by the
482 * number of readers before waking any processes up.
484 * This is an adaptation of the phase-fair R/W locks where at the
485 * reader phase (first waiter is a reader), all readers are eligible
486 * to acquire the lock at the same time irrespective of their order
487 * in the queue. The writers acquire the lock according to their
488 * order in the queue.
490 * We have to do wakeup in 2 passes to prevent the possibility that
491 * the reader count may be decremented before it is incremented. It
492 * is because the to-be-woken waiter may not have slept yet. So it
493 * may see waiter->task got cleared, finish its critical section and
494 * do an unlock before the reader count increment.
496 * 1) Collect the read-waiters in a separate list, count them and
497 * fully increment the reader count in rwsem.
498 * 2) For each waiters in the new list, clear waiter->task and
499 * put them into wake_q to be woken up later.
501 INIT_LIST_HEAD(&wlist);
502 list_for_each_entry_safe(waiter, tmp, &sem->wait_list, list) {
503 if (waiter->type == RWSEM_WAITING_FOR_WRITE)
507 list_move_tail(&waiter->list, &wlist);
510 * Limit # of readers that can be woken up per wakeup call.
512 if (woken >= MAX_READERS_WAKEUP)
516 adjustment = woken * RWSEM_READER_BIAS - adjustment;
517 lockevent_cond_inc(rwsem_wake_reader, woken);
519 oldcount = atomic_long_read(&sem->count);
520 if (list_empty(&sem->wait_list)) {
522 * Combined with list_move_tail() above, this implies
523 * rwsem_del_waiter().
525 adjustment -= RWSEM_FLAG_WAITERS;
526 if (oldcount & RWSEM_FLAG_HANDOFF)
527 adjustment -= RWSEM_FLAG_HANDOFF;
530 * When we've woken a reader, we no longer need to force
531 * writers to give up the lock and we can clear HANDOFF.
533 if (oldcount & RWSEM_FLAG_HANDOFF)
534 adjustment -= RWSEM_FLAG_HANDOFF;
538 atomic_long_add(adjustment, &sem->count);
541 list_for_each_entry_safe(waiter, tmp, &wlist, list) {
542 struct task_struct *tsk;
545 get_task_struct(tsk);
548 * Ensure calling get_task_struct() before setting the reader
549 * waiter to nil such that rwsem_down_read_slowpath() cannot
550 * race with do_exit() by always holding a reference count
551 * to the task to wakeup.
553 smp_store_release(&waiter->task, NULL);
555 * Ensure issuing the wakeup (either by us or someone else)
556 * after setting the reader waiter to nil.
558 wake_q_add_safe(wake_q, tsk);
563 * This function must be called with the sem->wait_lock held to prevent
564 * race conditions between checking the rwsem wait list and setting the
565 * sem->count accordingly.
567 * Implies rwsem_del_waiter() on success.
569 static inline bool rwsem_try_write_lock(struct rw_semaphore *sem,
570 struct rwsem_waiter *waiter)
572 struct rwsem_waiter *first = rwsem_first_waiter(sem);
575 lockdep_assert_held(&sem->wait_lock);
577 count = atomic_long_read(&sem->count);
579 bool has_handoff = !!(count & RWSEM_FLAG_HANDOFF);
583 * Honor handoff bit and yield only when the first
584 * waiter is the one that set it. Otherwisee, we
585 * still try to acquire the rwsem.
587 if (first->handoff_set && (waiter != first))
591 * First waiter can inherit a previously set handoff
592 * bit and spin on rwsem if lock acquisition fails.
595 waiter->handoff_set = true;
600 if (count & RWSEM_LOCK_MASK) {
601 if (has_handoff || (!rt_task(waiter->task) &&
602 !time_after(jiffies, waiter->timeout)))
605 new |= RWSEM_FLAG_HANDOFF;
607 new |= RWSEM_WRITER_LOCKED;
608 new &= ~RWSEM_FLAG_HANDOFF;
610 if (list_is_singular(&sem->wait_list))
611 new &= ~RWSEM_FLAG_WAITERS;
613 } while (!atomic_long_try_cmpxchg_acquire(&sem->count, &count, new));
616 * We have either acquired the lock with handoff bit cleared or
617 * set the handoff bit.
619 if (new & RWSEM_FLAG_HANDOFF) {
620 waiter->handoff_set = true;
621 lockevent_inc(rwsem_wlock_handoff);
626 * Have rwsem_try_write_lock() fully imply rwsem_del_waiter() on
629 list_del(&waiter->list);
630 rwsem_set_owner(sem);
635 * The rwsem_spin_on_owner() function returns the following 4 values
636 * depending on the lock owner state.
637 * OWNER_NULL : owner is currently NULL
638 * OWNER_WRITER: when owner changes and is a writer
639 * OWNER_READER: when owner changes and the new owner may be a reader.
640 * OWNER_NONSPINNABLE:
641 * when optimistic spinning has to stop because either the
642 * owner stops running, is unknown, or its timeslice has
647 OWNER_WRITER = 1 << 1,
648 OWNER_READER = 1 << 2,
649 OWNER_NONSPINNABLE = 1 << 3,
652 #ifdef CONFIG_RWSEM_SPIN_ON_OWNER
654 * Try to acquire write lock before the writer has been put on wait queue.
656 static inline bool rwsem_try_write_lock_unqueued(struct rw_semaphore *sem)
658 long count = atomic_long_read(&sem->count);
660 while (!(count & (RWSEM_LOCK_MASK|RWSEM_FLAG_HANDOFF))) {
661 if (atomic_long_try_cmpxchg_acquire(&sem->count, &count,
662 count | RWSEM_WRITER_LOCKED)) {
663 rwsem_set_owner(sem);
664 lockevent_inc(rwsem_opt_lock);
671 static inline bool owner_on_cpu(struct task_struct *owner)
674 * As lock holder preemption issue, we both skip spinning if
675 * task is not on cpu or its cpu is preempted
677 return owner->on_cpu && !vcpu_is_preempted(task_cpu(owner));
680 static inline bool rwsem_can_spin_on_owner(struct rw_semaphore *sem)
682 struct task_struct *owner;
686 if (need_resched()) {
687 lockevent_inc(rwsem_opt_fail);
693 owner = rwsem_owner_flags(sem, &flags);
695 * Don't check the read-owner as the entry may be stale.
697 if ((flags & RWSEM_NONSPINNABLE) ||
698 (owner && !(flags & RWSEM_READER_OWNED) && !owner_on_cpu(owner)))
703 lockevent_cond_inc(rwsem_opt_fail, !ret);
707 #define OWNER_SPINNABLE (OWNER_NULL | OWNER_WRITER | OWNER_READER)
709 static inline enum owner_state
710 rwsem_owner_state(struct task_struct *owner, unsigned long flags)
712 if (flags & RWSEM_NONSPINNABLE)
713 return OWNER_NONSPINNABLE;
715 if (flags & RWSEM_READER_OWNED)
718 return owner ? OWNER_WRITER : OWNER_NULL;
721 static noinline enum owner_state
722 rwsem_spin_on_owner(struct rw_semaphore *sem)
724 struct task_struct *new, *owner;
725 unsigned long flags, new_flags;
726 enum owner_state state;
728 owner = rwsem_owner_flags(sem, &flags);
729 state = rwsem_owner_state(owner, flags);
730 if (state != OWNER_WRITER)
736 * When a waiting writer set the handoff flag, it may spin
737 * on the owner as well. Once that writer acquires the lock,
738 * we can spin on it. So we don't need to quit even when the
739 * handoff bit is set.
741 new = rwsem_owner_flags(sem, &new_flags);
742 if ((new != owner) || (new_flags != flags)) {
743 state = rwsem_owner_state(new, new_flags);
748 * Ensure we emit the owner->on_cpu, dereference _after_
749 * checking sem->owner still matches owner, if that fails,
750 * owner might point to free()d memory, if it still matches,
751 * the rcu_read_lock() ensures the memory stays valid.
755 if (need_resched() || !owner_on_cpu(owner)) {
756 state = OWNER_NONSPINNABLE;
768 * Calculate reader-owned rwsem spinning threshold for writer
770 * The more readers own the rwsem, the longer it will take for them to
771 * wind down and free the rwsem. So the empirical formula used to
772 * determine the actual spinning time limit here is:
774 * Spinning threshold = (10 + nr_readers/2)us
776 * The limit is capped to a maximum of 25us (30 readers). This is just
777 * a heuristic and is subjected to change in the future.
779 static inline u64 rwsem_rspin_threshold(struct rw_semaphore *sem)
781 long count = atomic_long_read(&sem->count);
782 int readers = count >> RWSEM_READER_SHIFT;
787 delta = (20 + readers) * NSEC_PER_USEC / 2;
789 return sched_clock() + delta;
792 static bool rwsem_optimistic_spin(struct rw_semaphore *sem)
795 int prev_owner_state = OWNER_NULL;
797 u64 rspin_threshold = 0;
801 /* sem->wait_lock should not be held when doing optimistic spinning */
802 if (!osq_lock(&sem->osq))
806 * Optimistically spin on the owner field and attempt to acquire the
807 * lock whenever the owner changes. Spinning will be stopped when:
808 * 1) the owning writer isn't running; or
809 * 2) readers own the lock and spinning time has exceeded limit.
812 enum owner_state owner_state;
814 owner_state = rwsem_spin_on_owner(sem);
815 if (!(owner_state & OWNER_SPINNABLE))
819 * Try to acquire the lock
821 taken = rwsem_try_write_lock_unqueued(sem);
827 * Time-based reader-owned rwsem optimistic spinning
829 if (owner_state == OWNER_READER) {
831 * Re-initialize rspin_threshold every time when
832 * the owner state changes from non-reader to reader.
833 * This allows a writer to steal the lock in between
834 * 2 reader phases and have the threshold reset at
835 * the beginning of the 2nd reader phase.
837 if (prev_owner_state != OWNER_READER) {
838 if (rwsem_test_oflags(sem, RWSEM_NONSPINNABLE))
840 rspin_threshold = rwsem_rspin_threshold(sem);
845 * Check time threshold once every 16 iterations to
846 * avoid calling sched_clock() too frequently so
847 * as to reduce the average latency between the times
848 * when the lock becomes free and when the spinner
849 * is ready to do a trylock.
851 else if (!(++loop & 0xf) && (sched_clock() > rspin_threshold)) {
852 rwsem_set_nonspinnable(sem);
853 lockevent_inc(rwsem_opt_nospin);
859 * An RT task cannot do optimistic spinning if it cannot
860 * be sure the lock holder is running or live-lock may
861 * happen if the current task and the lock holder happen
862 * to run in the same CPU. However, aborting optimistic
863 * spinning while a NULL owner is detected may miss some
864 * opportunity where spinning can continue without causing
867 * There are 2 possible cases where an RT task may be able
868 * to continue spinning.
870 * 1) The lock owner is in the process of releasing the
871 * lock, sem->owner is cleared but the lock has not
873 * 2) The lock was free and owner cleared, but another
874 * task just comes in and acquire the lock before
875 * we try to get it. The new owner may be a spinnable
878 * To take advantage of two scenarios listed above, the RT
879 * task is made to retry one more time to see if it can
880 * acquire the lock or continue spinning on the new owning
881 * writer. Of course, if the time lag is long enough or the
882 * new owner is not a writer or spinnable, the RT task will
885 * If the owner is a writer, the need_resched() check is
886 * done inside rwsem_spin_on_owner(). If the owner is not
887 * a writer, need_resched() check needs to be done here.
889 if (owner_state != OWNER_WRITER) {
892 if (rt_task(current) &&
893 (prev_owner_state != OWNER_WRITER))
896 prev_owner_state = owner_state;
899 * The cpu_relax() call is a compiler barrier which forces
900 * everything in this loop to be re-loaded. We don't need
901 * memory barriers as we'll eventually observe the right
902 * values at the cost of a few extra spins.
906 osq_unlock(&sem->osq);
909 lockevent_cond_inc(rwsem_opt_fail, !taken);
914 * Clear the owner's RWSEM_NONSPINNABLE bit if it is set. This should
915 * only be called when the reader count reaches 0.
917 static inline void clear_nonspinnable(struct rw_semaphore *sem)
919 if (rwsem_test_oflags(sem, RWSEM_NONSPINNABLE))
920 atomic_long_andnot(RWSEM_NONSPINNABLE, &sem->owner);
924 static inline bool rwsem_can_spin_on_owner(struct rw_semaphore *sem)
929 static inline bool rwsem_optimistic_spin(struct rw_semaphore *sem)
934 static inline void clear_nonspinnable(struct rw_semaphore *sem) { }
936 static inline enum owner_state
937 rwsem_spin_on_owner(struct rw_semaphore *sem)
939 return OWNER_NONSPINNABLE;
944 * Wait for the read lock to be granted
946 static struct rw_semaphore __sched *
947 rwsem_down_read_slowpath(struct rw_semaphore *sem, long count, unsigned int state)
949 long adjustment = -RWSEM_READER_BIAS;
950 long rcnt = (count >> RWSEM_READER_SHIFT);
951 struct rwsem_waiter waiter;
952 DEFINE_WAKE_Q(wake_q);
956 * To prevent a constant stream of readers from starving a sleeping
957 * waiter, don't attempt optimistic lock stealing if the lock is
958 * currently owned by readers.
960 if ((atomic_long_read(&sem->owner) & RWSEM_READER_OWNED) &&
961 (rcnt > 1) && !(count & RWSEM_WRITER_LOCKED))
965 * Reader optimistic lock stealing.
967 if (!(count & (RWSEM_WRITER_LOCKED | RWSEM_FLAG_HANDOFF))) {
968 rwsem_set_reader_owned(sem);
969 lockevent_inc(rwsem_rlock_steal);
972 * Wake up other readers in the wait queue if it is
975 if ((rcnt == 1) && (count & RWSEM_FLAG_WAITERS)) {
976 raw_spin_lock_irq(&sem->wait_lock);
977 if (!list_empty(&sem->wait_list))
978 rwsem_mark_wake(sem, RWSEM_WAKE_READ_OWNED,
980 raw_spin_unlock_irq(&sem->wait_lock);
987 waiter.task = current;
988 waiter.type = RWSEM_WAITING_FOR_READ;
989 waiter.timeout = jiffies + RWSEM_WAIT_TIMEOUT;
990 waiter.handoff_set = false;
992 raw_spin_lock_irq(&sem->wait_lock);
993 if (list_empty(&sem->wait_list)) {
995 * In case the wait queue is empty and the lock isn't owned
996 * by a writer or has the handoff bit set, this reader can
997 * exit the slowpath and return immediately as its
998 * RWSEM_READER_BIAS has already been set in the count.
1000 if (!(atomic_long_read(&sem->count) &
1001 (RWSEM_WRITER_MASK | RWSEM_FLAG_HANDOFF))) {
1002 /* Provide lock ACQUIRE */
1003 smp_acquire__after_ctrl_dep();
1004 raw_spin_unlock_irq(&sem->wait_lock);
1005 rwsem_set_reader_owned(sem);
1006 lockevent_inc(rwsem_rlock_fast);
1009 adjustment += RWSEM_FLAG_WAITERS;
1011 rwsem_add_waiter(sem, &waiter);
1013 /* we're now waiting on the lock, but no longer actively locking */
1014 count = atomic_long_add_return(adjustment, &sem->count);
1017 * If there are no active locks, wake the front queued process(es).
1019 * If there are no writers and we are first in the queue,
1020 * wake our own waiter to join the existing active readers !
1022 if (!(count & RWSEM_LOCK_MASK)) {
1023 clear_nonspinnable(sem);
1026 if (wake || (!(count & RWSEM_WRITER_MASK) &&
1027 (adjustment & RWSEM_FLAG_WAITERS)))
1028 rwsem_mark_wake(sem, RWSEM_WAKE_ANY, &wake_q);
1030 raw_spin_unlock_irq(&sem->wait_lock);
1033 /* wait to be given the lock */
1035 set_current_state(state);
1036 if (!smp_load_acquire(&waiter.task)) {
1037 /* Matches rwsem_mark_wake()'s smp_store_release(). */
1040 if (signal_pending_state(state, current)) {
1041 raw_spin_lock_irq(&sem->wait_lock);
1044 raw_spin_unlock_irq(&sem->wait_lock);
1045 /* Ordered by sem->wait_lock against rwsem_mark_wake(). */
1049 lockevent_inc(rwsem_sleep_reader);
1052 __set_current_state(TASK_RUNNING);
1053 lockevent_inc(rwsem_rlock);
1057 rwsem_del_waiter(sem, &waiter);
1058 raw_spin_unlock_irq(&sem->wait_lock);
1059 __set_current_state(TASK_RUNNING);
1060 lockevent_inc(rwsem_rlock_fail);
1061 return ERR_PTR(-EINTR);
1065 * Wait until we successfully acquire the write lock
1067 static struct rw_semaphore *
1068 rwsem_down_write_slowpath(struct rw_semaphore *sem, int state)
1071 struct rwsem_waiter waiter;
1072 DEFINE_WAKE_Q(wake_q);
1074 /* do optimistic spinning and steal lock if possible */
1075 if (rwsem_can_spin_on_owner(sem) && rwsem_optimistic_spin(sem)) {
1076 /* rwsem_optimistic_spin() implies ACQUIRE on success */
1081 * Optimistic spinning failed, proceed to the slowpath
1082 * and block until we can acquire the sem.
1084 waiter.task = current;
1085 waiter.type = RWSEM_WAITING_FOR_WRITE;
1086 waiter.timeout = jiffies + RWSEM_WAIT_TIMEOUT;
1087 waiter.handoff_set = false;
1089 raw_spin_lock_irq(&sem->wait_lock);
1090 rwsem_add_waiter(sem, &waiter);
1092 /* we're now waiting on the lock */
1093 if (rwsem_first_waiter(sem) != &waiter) {
1094 count = atomic_long_read(&sem->count);
1097 * If there were already threads queued before us and:
1098 * 1) there are no active locks, wake the front
1099 * queued process(es) as the handoff bit might be set.
1100 * 2) there are no active writers and some readers, the lock
1101 * must be read owned; so we try to wake any read lock
1102 * waiters that were queued ahead of us.
1104 if (count & RWSEM_WRITER_MASK)
1107 rwsem_mark_wake(sem, (count & RWSEM_READER_MASK)
1108 ? RWSEM_WAKE_READERS
1109 : RWSEM_WAKE_ANY, &wake_q);
1111 if (!wake_q_empty(&wake_q)) {
1113 * We want to minimize wait_lock hold time especially
1114 * when a large number of readers are to be woken up.
1116 raw_spin_unlock_irq(&sem->wait_lock);
1118 wake_q_init(&wake_q); /* Used again, reinit */
1119 raw_spin_lock_irq(&sem->wait_lock);
1122 atomic_long_or(RWSEM_FLAG_WAITERS, &sem->count);
1126 /* wait until we successfully acquire the lock */
1127 set_current_state(state);
1129 if (rwsem_try_write_lock(sem, &waiter)) {
1130 /* rwsem_try_write_lock() implies ACQUIRE on success */
1134 raw_spin_unlock_irq(&sem->wait_lock);
1136 if (signal_pending_state(state, current))
1140 * After setting the handoff bit and failing to acquire
1141 * the lock, attempt to spin on owner to accelerate lock
1142 * transfer. If the previous owner is a on-cpu writer and it
1143 * has just released the lock, OWNER_NULL will be returned.
1144 * In this case, we attempt to acquire the lock again
1147 if (waiter.handoff_set) {
1148 enum owner_state owner_state;
1151 owner_state = rwsem_spin_on_owner(sem);
1154 if (owner_state == OWNER_NULL)
1159 lockevent_inc(rwsem_sleep_writer);
1160 set_current_state(state);
1162 raw_spin_lock_irq(&sem->wait_lock);
1164 __set_current_state(TASK_RUNNING);
1165 raw_spin_unlock_irq(&sem->wait_lock);
1166 lockevent_inc(rwsem_wlock);
1170 __set_current_state(TASK_RUNNING);
1171 raw_spin_lock_irq(&sem->wait_lock);
1172 rwsem_del_waiter(sem, &waiter);
1173 if (!list_empty(&sem->wait_list))
1174 rwsem_mark_wake(sem, RWSEM_WAKE_ANY, &wake_q);
1175 raw_spin_unlock_irq(&sem->wait_lock);
1177 lockevent_inc(rwsem_wlock_fail);
1178 return ERR_PTR(-EINTR);
1182 * handle waking up a waiter on the semaphore
1183 * - up_read/up_write has decremented the active part of count if we come here
1185 static struct rw_semaphore *rwsem_wake(struct rw_semaphore *sem)
1187 unsigned long flags;
1188 DEFINE_WAKE_Q(wake_q);
1190 raw_spin_lock_irqsave(&sem->wait_lock, flags);
1192 if (!list_empty(&sem->wait_list))
1193 rwsem_mark_wake(sem, RWSEM_WAKE_ANY, &wake_q);
1195 raw_spin_unlock_irqrestore(&sem->wait_lock, flags);
1202 * downgrade a write lock into a read lock
1203 * - caller incremented waiting part of count and discovered it still negative
1204 * - just wake up any readers at the front of the queue
1206 static struct rw_semaphore *rwsem_downgrade_wake(struct rw_semaphore *sem)
1208 unsigned long flags;
1209 DEFINE_WAKE_Q(wake_q);
1211 raw_spin_lock_irqsave(&sem->wait_lock, flags);
1213 if (!list_empty(&sem->wait_list))
1214 rwsem_mark_wake(sem, RWSEM_WAKE_READ_OWNED, &wake_q);
1216 raw_spin_unlock_irqrestore(&sem->wait_lock, flags);
1225 static inline int __down_read_common(struct rw_semaphore *sem, int state)
1229 if (!rwsem_read_trylock(sem, &count)) {
1230 if (IS_ERR(rwsem_down_read_slowpath(sem, count, state)))
1232 DEBUG_RWSEMS_WARN_ON(!is_rwsem_reader_owned(sem), sem);
1237 static inline void __down_read(struct rw_semaphore *sem)
1239 __down_read_common(sem, TASK_UNINTERRUPTIBLE);
1242 static inline int __down_read_interruptible(struct rw_semaphore *sem)
1244 return __down_read_common(sem, TASK_INTERRUPTIBLE);
1247 static inline int __down_read_killable(struct rw_semaphore *sem)
1249 return __down_read_common(sem, TASK_KILLABLE);
1252 static inline int __down_read_trylock(struct rw_semaphore *sem)
1256 DEBUG_RWSEMS_WARN_ON(sem->magic != sem, sem);
1259 * Optimize for the case when the rwsem is not locked at all.
1261 tmp = RWSEM_UNLOCKED_VALUE;
1263 if (atomic_long_try_cmpxchg_acquire(&sem->count, &tmp,
1264 tmp + RWSEM_READER_BIAS)) {
1265 rwsem_set_reader_owned(sem);
1268 } while (!(tmp & RWSEM_READ_FAILED_MASK));
1275 static inline int __down_write_common(struct rw_semaphore *sem, int state)
1277 if (unlikely(!rwsem_write_trylock(sem))) {
1278 if (IS_ERR(rwsem_down_write_slowpath(sem, state)))
1285 static inline void __down_write(struct rw_semaphore *sem)
1287 __down_write_common(sem, TASK_UNINTERRUPTIBLE);
1290 static inline int __down_write_killable(struct rw_semaphore *sem)
1292 return __down_write_common(sem, TASK_KILLABLE);
1295 static inline int __down_write_trylock(struct rw_semaphore *sem)
1297 DEBUG_RWSEMS_WARN_ON(sem->magic != sem, sem);
1298 return rwsem_write_trylock(sem);
1302 * unlock after reading
1304 static inline void __up_read(struct rw_semaphore *sem)
1308 DEBUG_RWSEMS_WARN_ON(sem->magic != sem, sem);
1309 DEBUG_RWSEMS_WARN_ON(!is_rwsem_reader_owned(sem), sem);
1311 rwsem_clear_reader_owned(sem);
1312 tmp = atomic_long_add_return_release(-RWSEM_READER_BIAS, &sem->count);
1313 DEBUG_RWSEMS_WARN_ON(tmp < 0, sem);
1314 if (unlikely((tmp & (RWSEM_LOCK_MASK|RWSEM_FLAG_WAITERS)) ==
1315 RWSEM_FLAG_WAITERS)) {
1316 clear_nonspinnable(sem);
1322 * unlock after writing
1324 static inline void __up_write(struct rw_semaphore *sem)
1328 DEBUG_RWSEMS_WARN_ON(sem->magic != sem, sem);
1330 * sem->owner may differ from current if the ownership is transferred
1331 * to an anonymous writer by setting the RWSEM_NONSPINNABLE bits.
1333 DEBUG_RWSEMS_WARN_ON((rwsem_owner(sem) != current) &&
1334 !rwsem_test_oflags(sem, RWSEM_NONSPINNABLE), sem);
1336 rwsem_clear_owner(sem);
1337 tmp = atomic_long_fetch_add_release(-RWSEM_WRITER_LOCKED, &sem->count);
1338 if (unlikely(tmp & RWSEM_FLAG_WAITERS))
1343 * downgrade write lock to read lock
1345 static inline void __downgrade_write(struct rw_semaphore *sem)
1350 * When downgrading from exclusive to shared ownership,
1351 * anything inside the write-locked region cannot leak
1352 * into the read side. In contrast, anything in the
1353 * read-locked region is ok to be re-ordered into the
1354 * write side. As such, rely on RELEASE semantics.
1356 DEBUG_RWSEMS_WARN_ON(rwsem_owner(sem) != current, sem);
1357 tmp = atomic_long_fetch_add_release(
1358 -RWSEM_WRITER_LOCKED+RWSEM_READER_BIAS, &sem->count);
1359 rwsem_set_reader_owned(sem);
1360 if (tmp & RWSEM_FLAG_WAITERS)
1361 rwsem_downgrade_wake(sem);
1364 #else /* !CONFIG_PREEMPT_RT */
1366 #define RT_MUTEX_BUILD_MUTEX
1367 #include "rtmutex.c"
1369 #define rwbase_set_and_save_current_state(state) \
1370 set_current_state(state)
1372 #define rwbase_restore_current_state() \
1373 __set_current_state(TASK_RUNNING)
1375 #define rwbase_rtmutex_lock_state(rtm, state) \
1376 __rt_mutex_lock(rtm, state)
1378 #define rwbase_rtmutex_slowlock_locked(rtm, state) \
1379 __rt_mutex_slowlock_locked(rtm, NULL, state)
1381 #define rwbase_rtmutex_unlock(rtm) \
1382 __rt_mutex_unlock(rtm)
1384 #define rwbase_rtmutex_trylock(rtm) \
1385 __rt_mutex_trylock(rtm)
1387 #define rwbase_signal_pending_state(state, current) \
1388 signal_pending_state(state, current)
1390 #define rwbase_schedule() \
1393 #include "rwbase_rt.c"
1395 void __init_rwsem(struct rw_semaphore *sem, const char *name,
1396 struct lock_class_key *key)
1398 init_rwbase_rt(&(sem)->rwbase);
1400 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1401 debug_check_no_locks_freed((void *)sem, sizeof(*sem));
1402 lockdep_init_map_wait(&sem->dep_map, name, key, 0, LD_WAIT_SLEEP);
1405 EXPORT_SYMBOL(__init_rwsem);
1407 static inline void __down_read(struct rw_semaphore *sem)
1409 rwbase_read_lock(&sem->rwbase, TASK_UNINTERRUPTIBLE);
1412 static inline int __down_read_interruptible(struct rw_semaphore *sem)
1414 return rwbase_read_lock(&sem->rwbase, TASK_INTERRUPTIBLE);
1417 static inline int __down_read_killable(struct rw_semaphore *sem)
1419 return rwbase_read_lock(&sem->rwbase, TASK_KILLABLE);
1422 static inline int __down_read_trylock(struct rw_semaphore *sem)
1424 return rwbase_read_trylock(&sem->rwbase);
1427 static inline void __up_read(struct rw_semaphore *sem)
1429 rwbase_read_unlock(&sem->rwbase, TASK_NORMAL);
1432 static inline void __sched __down_write(struct rw_semaphore *sem)
1434 rwbase_write_lock(&sem->rwbase, TASK_UNINTERRUPTIBLE);
1437 static inline int __sched __down_write_killable(struct rw_semaphore *sem)
1439 return rwbase_write_lock(&sem->rwbase, TASK_KILLABLE);
1442 static inline int __down_write_trylock(struct rw_semaphore *sem)
1444 return rwbase_write_trylock(&sem->rwbase);
1447 static inline void __up_write(struct rw_semaphore *sem)
1449 rwbase_write_unlock(&sem->rwbase);
1452 static inline void __downgrade_write(struct rw_semaphore *sem)
1454 rwbase_write_downgrade(&sem->rwbase);
1457 /* Debug stubs for the common API */
1458 #define DEBUG_RWSEMS_WARN_ON(c, sem)
1460 static inline void __rwsem_set_reader_owned(struct rw_semaphore *sem,
1461 struct task_struct *owner)
1465 static inline bool is_rwsem_reader_owned(struct rw_semaphore *sem)
1467 int count = atomic_read(&sem->rwbase.readers);
1469 return count < 0 && count != READER_BIAS;
1472 #endif /* CONFIG_PREEMPT_RT */
1477 void __sched down_read(struct rw_semaphore *sem)
1480 rwsem_acquire_read(&sem->dep_map, 0, 0, _RET_IP_);
1482 LOCK_CONTENDED(sem, __down_read_trylock, __down_read);
1484 EXPORT_SYMBOL(down_read);
1486 int __sched down_read_interruptible(struct rw_semaphore *sem)
1489 rwsem_acquire_read(&sem->dep_map, 0, 0, _RET_IP_);
1491 if (LOCK_CONTENDED_RETURN(sem, __down_read_trylock, __down_read_interruptible)) {
1492 rwsem_release(&sem->dep_map, _RET_IP_);
1498 EXPORT_SYMBOL(down_read_interruptible);
1500 int __sched down_read_killable(struct rw_semaphore *sem)
1503 rwsem_acquire_read(&sem->dep_map, 0, 0, _RET_IP_);
1505 if (LOCK_CONTENDED_RETURN(sem, __down_read_trylock, __down_read_killable)) {
1506 rwsem_release(&sem->dep_map, _RET_IP_);
1512 EXPORT_SYMBOL(down_read_killable);
1515 * trylock for reading -- returns 1 if successful, 0 if contention
1517 int down_read_trylock(struct rw_semaphore *sem)
1519 int ret = __down_read_trylock(sem);
1522 rwsem_acquire_read(&sem->dep_map, 0, 1, _RET_IP_);
1525 EXPORT_SYMBOL(down_read_trylock);
1530 void __sched down_write(struct rw_semaphore *sem)
1533 rwsem_acquire(&sem->dep_map, 0, 0, _RET_IP_);
1534 LOCK_CONTENDED(sem, __down_write_trylock, __down_write);
1536 EXPORT_SYMBOL(down_write);
1541 int __sched down_write_killable(struct rw_semaphore *sem)
1544 rwsem_acquire(&sem->dep_map, 0, 0, _RET_IP_);
1546 if (LOCK_CONTENDED_RETURN(sem, __down_write_trylock,
1547 __down_write_killable)) {
1548 rwsem_release(&sem->dep_map, _RET_IP_);
1554 EXPORT_SYMBOL(down_write_killable);
1557 * trylock for writing -- returns 1 if successful, 0 if contention
1559 int down_write_trylock(struct rw_semaphore *sem)
1561 int ret = __down_write_trylock(sem);
1564 rwsem_acquire(&sem->dep_map, 0, 1, _RET_IP_);
1568 EXPORT_SYMBOL(down_write_trylock);
1571 * release a read lock
1573 void up_read(struct rw_semaphore *sem)
1575 rwsem_release(&sem->dep_map, _RET_IP_);
1578 EXPORT_SYMBOL(up_read);
1581 * release a write lock
1583 void up_write(struct rw_semaphore *sem)
1585 rwsem_release(&sem->dep_map, _RET_IP_);
1588 EXPORT_SYMBOL(up_write);
1591 * downgrade write lock to read lock
1593 void downgrade_write(struct rw_semaphore *sem)
1595 lock_downgrade(&sem->dep_map, _RET_IP_);
1596 __downgrade_write(sem);
1598 EXPORT_SYMBOL(downgrade_write);
1600 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1602 void down_read_nested(struct rw_semaphore *sem, int subclass)
1605 rwsem_acquire_read(&sem->dep_map, subclass, 0, _RET_IP_);
1606 LOCK_CONTENDED(sem, __down_read_trylock, __down_read);
1608 EXPORT_SYMBOL(down_read_nested);
1610 int down_read_killable_nested(struct rw_semaphore *sem, int subclass)
1613 rwsem_acquire_read(&sem->dep_map, subclass, 0, _RET_IP_);
1615 if (LOCK_CONTENDED_RETURN(sem, __down_read_trylock, __down_read_killable)) {
1616 rwsem_release(&sem->dep_map, _RET_IP_);
1622 EXPORT_SYMBOL(down_read_killable_nested);
1624 void _down_write_nest_lock(struct rw_semaphore *sem, struct lockdep_map *nest)
1627 rwsem_acquire_nest(&sem->dep_map, 0, 0, nest, _RET_IP_);
1628 LOCK_CONTENDED(sem, __down_write_trylock, __down_write);
1630 EXPORT_SYMBOL(_down_write_nest_lock);
1632 void down_read_non_owner(struct rw_semaphore *sem)
1636 __rwsem_set_reader_owned(sem, NULL);
1638 EXPORT_SYMBOL(down_read_non_owner);
1640 void down_write_nested(struct rw_semaphore *sem, int subclass)
1643 rwsem_acquire(&sem->dep_map, subclass, 0, _RET_IP_);
1644 LOCK_CONTENDED(sem, __down_write_trylock, __down_write);
1646 EXPORT_SYMBOL(down_write_nested);
1648 int __sched down_write_killable_nested(struct rw_semaphore *sem, int subclass)
1651 rwsem_acquire(&sem->dep_map, subclass, 0, _RET_IP_);
1653 if (LOCK_CONTENDED_RETURN(sem, __down_write_trylock,
1654 __down_write_killable)) {
1655 rwsem_release(&sem->dep_map, _RET_IP_);
1661 EXPORT_SYMBOL(down_write_killable_nested);
1663 void up_read_non_owner(struct rw_semaphore *sem)
1665 DEBUG_RWSEMS_WARN_ON(!is_rwsem_reader_owned(sem), sem);
1668 EXPORT_SYMBOL(up_read_non_owner);