2 * Fast Userspace Mutexes (which I call "Futexes!").
3 * (C) Rusty Russell, IBM 2002
5 * Generalized futexes, futex requeueing, misc fixes by Ingo Molnar
6 * (C) Copyright 2003 Red Hat Inc, All Rights Reserved
8 * Removed page pinning, fix privately mapped COW pages and other cleanups
9 * (C) Copyright 2003, 2004 Jamie Lokier
11 * Robust futex support started by Ingo Molnar
12 * (C) Copyright 2006 Red Hat Inc, All Rights Reserved
13 * Thanks to Thomas Gleixner for suggestions, analysis and fixes.
15 * PI-futex support started by Ingo Molnar and Thomas Gleixner
16 * Copyright (C) 2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
17 * Copyright (C) 2006 Timesys Corp., Thomas Gleixner <tglx@timesys.com>
19 * PRIVATE futexes by Eric Dumazet
20 * Copyright (C) 2007 Eric Dumazet <dada1@cosmosbay.com>
22 * Thanks to Ben LaHaise for yelling "hashed waitqueues" loudly
23 * enough at me, Linus for the original (flawed) idea, Matthew
24 * Kirkwood for proof-of-concept implementation.
26 * "The futexes are also cursed."
27 * "But they come in a choice of three flavours!"
29 * This program is free software; you can redistribute it and/or modify
30 * it under the terms of the GNU General Public License as published by
31 * the Free Software Foundation; either version 2 of the License, or
32 * (at your option) any later version.
34 * This program is distributed in the hope that it will be useful,
35 * but WITHOUT ANY WARRANTY; without even the implied warranty of
36 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
37 * GNU General Public License for more details.
39 * You should have received a copy of the GNU General Public License
40 * along with this program; if not, write to the Free Software
41 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
43 #include <linux/slab.h>
44 #include <linux/poll.h>
46 #include <linux/file.h>
47 #include <linux/jhash.h>
48 #include <linux/init.h>
49 #include <linux/futex.h>
50 #include <linux/mount.h>
51 #include <linux/pagemap.h>
52 #include <linux/syscalls.h>
53 #include <linux/signal.h>
54 #include <linux/module.h>
55 #include <linux/magic.h>
56 #include <linux/pid.h>
57 #include <linux/nsproxy.h>
59 #include <asm/futex.h>
61 #include "rtmutex_common.h"
63 int __read_mostly futex_cmpxchg_enabled;
65 #define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8)
68 * Priority Inheritance state:
70 struct futex_pi_state {
72 * list of 'owned' pi_state instances - these have to be
73 * cleaned up in do_exit() if the task exits prematurely:
75 struct list_head list;
80 struct rt_mutex pi_mutex;
82 struct task_struct *owner;
89 * We use this hashed waitqueue instead of a normal wait_queue_t, so
90 * we can wake only the relevant ones (hashed queues may be shared).
92 * A futex_q has a woken state, just like tasks have TASK_RUNNING.
93 * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0.
94 * The order of wakup is always to make the first condition true, then
95 * wake up q->waiter, then make the second condition true.
98 struct plist_node list;
99 /* There can only be a single waiter */
100 wait_queue_head_t waiter;
102 /* Which hash list lock to use: */
103 spinlock_t *lock_ptr;
105 /* Key which the futex is hashed on: */
108 /* Optional priority inheritance state: */
109 struct futex_pi_state *pi_state;
110 struct task_struct *task;
112 /* Bitset for the optional bitmasked wakeup */
117 * Split the global futex_lock into every hash list lock.
119 struct futex_hash_bucket {
121 struct plist_head chain;
124 static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS];
127 * We hash on the keys returned from get_futex_key (see below).
129 static struct futex_hash_bucket *hash_futex(union futex_key *key)
131 u32 hash = jhash2((u32*)&key->both.word,
132 (sizeof(key->both.word)+sizeof(key->both.ptr))/4,
134 return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)];
138 * Return 1 if two futex_keys are equal, 0 otherwise.
140 static inline int match_futex(union futex_key *key1, union futex_key *key2)
142 return (key1->both.word == key2->both.word
143 && key1->both.ptr == key2->both.ptr
144 && key1->both.offset == key2->both.offset);
148 * Take a reference to the resource addressed by a key.
149 * Can be called while holding spinlocks.
152 static void get_futex_key_refs(union futex_key *key)
157 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
159 atomic_inc(&key->shared.inode->i_count);
161 case FUT_OFF_MMSHARED:
162 atomic_inc(&key->private.mm->mm_count);
168 * Drop a reference to the resource addressed by a key.
169 * The hash bucket spinlock must not be held.
171 static void drop_futex_key_refs(union futex_key *key)
173 if (!key->both.ptr) {
174 /* If we're here then we tried to put a key we failed to get */
179 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
181 iput(key->shared.inode);
183 case FUT_OFF_MMSHARED:
184 mmdrop(key->private.mm);
190 * get_futex_key - Get parameters which are the keys for a futex.
191 * @uaddr: virtual address of the futex
192 * @shared: NULL for a PROCESS_PRIVATE futex,
193 * ¤t->mm->mmap_sem for a PROCESS_SHARED futex
194 * @key: address where result is stored.
196 * Returns a negative error code or 0
197 * The key words are stored in *key on success.
199 * For shared mappings, it's (page->index, vma->vm_file->f_path.dentry->d_inode,
200 * offset_within_page). For private mappings, it's (uaddr, current->mm).
201 * We can usually work out the index without swapping in the page.
203 * fshared is NULL for PROCESS_PRIVATE futexes
204 * For other futexes, it points to ¤t->mm->mmap_sem and
205 * caller must have taken the reader lock. but NOT any spinlocks.
207 static int get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key)
209 unsigned long address = (unsigned long)uaddr;
210 struct mm_struct *mm = current->mm;
215 * The futex address must be "naturally" aligned.
217 key->both.offset = address % PAGE_SIZE;
218 if (unlikely((address % sizeof(u32)) != 0))
220 address -= key->both.offset;
223 * PROCESS_PRIVATE futexes are fast.
224 * As the mm cannot disappear under us and the 'key' only needs
225 * virtual address, we dont even have to find the underlying vma.
226 * Note : We do have to check 'uaddr' is a valid user address,
227 * but access_ok() should be faster than find_vma()
230 if (unlikely(!access_ok(VERIFY_WRITE, uaddr, sizeof(u32))))
232 key->private.mm = mm;
233 key->private.address = address;
234 get_futex_key_refs(key);
239 err = get_user_pages_fast(address, 1, 0, &page);
244 if (!page->mapping) {
251 * Private mappings are handled in a simple way.
253 * NOTE: When userspace waits on a MAP_SHARED mapping, even if
254 * it's a read-only handle, it's expected that futexes attach to
255 * the object not the particular process.
257 if (PageAnon(page)) {
258 key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */
259 key->private.mm = mm;
260 key->private.address = address;
262 key->both.offset |= FUT_OFF_INODE; /* inode-based key */
263 key->shared.inode = page->mapping->host;
264 key->shared.pgoff = page->index;
267 get_futex_key_refs(key);
275 void put_futex_key(int fshared, union futex_key *key)
277 drop_futex_key_refs(key);
280 static u32 cmpxchg_futex_value_locked(u32 __user *uaddr, u32 uval, u32 newval)
285 curval = futex_atomic_cmpxchg_inatomic(uaddr, uval, newval);
291 static int get_futex_value_locked(u32 *dest, u32 __user *from)
296 ret = __copy_from_user_inatomic(dest, from, sizeof(u32));
299 return ret ? -EFAULT : 0;
305 static int futex_handle_fault(unsigned long address, int attempt)
307 struct vm_area_struct * vma;
308 struct mm_struct *mm = current->mm;
314 down_read(&mm->mmap_sem);
315 vma = find_vma(mm, address);
316 if (vma && address >= vma->vm_start &&
317 (vma->vm_flags & VM_WRITE)) {
319 fault = handle_mm_fault(mm, vma, address, 1);
320 if (unlikely((fault & VM_FAULT_ERROR))) {
322 /* XXX: let's do this when we verify it is OK */
323 if (ret & VM_FAULT_OOM)
328 if (fault & VM_FAULT_MAJOR)
334 up_read(&mm->mmap_sem);
341 static int refill_pi_state_cache(void)
343 struct futex_pi_state *pi_state;
345 if (likely(current->pi_state_cache))
348 pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL);
353 INIT_LIST_HEAD(&pi_state->list);
354 /* pi_mutex gets initialized later */
355 pi_state->owner = NULL;
356 atomic_set(&pi_state->refcount, 1);
357 pi_state->key = FUTEX_KEY_INIT;
359 current->pi_state_cache = pi_state;
364 static struct futex_pi_state * alloc_pi_state(void)
366 struct futex_pi_state *pi_state = current->pi_state_cache;
369 current->pi_state_cache = NULL;
374 static void free_pi_state(struct futex_pi_state *pi_state)
376 if (!atomic_dec_and_test(&pi_state->refcount))
380 * If pi_state->owner is NULL, the owner is most probably dying
381 * and has cleaned up the pi_state already
383 if (pi_state->owner) {
384 spin_lock_irq(&pi_state->owner->pi_lock);
385 list_del_init(&pi_state->list);
386 spin_unlock_irq(&pi_state->owner->pi_lock);
388 rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner);
391 if (current->pi_state_cache)
395 * pi_state->list is already empty.
396 * clear pi_state->owner.
397 * refcount is at 0 - put it back to 1.
399 pi_state->owner = NULL;
400 atomic_set(&pi_state->refcount, 1);
401 current->pi_state_cache = pi_state;
406 * Look up the task based on what TID userspace gave us.
409 static struct task_struct * futex_find_get_task(pid_t pid)
411 struct task_struct *p;
414 p = find_task_by_vpid(pid);
415 if (!p || ((current->euid != p->euid) && (current->euid != p->uid)))
426 * This task is holding PI mutexes at exit time => bad.
427 * Kernel cleans up PI-state, but userspace is likely hosed.
428 * (Robust-futex cleanup is separate and might save the day for userspace.)
430 void exit_pi_state_list(struct task_struct *curr)
432 struct list_head *next, *head = &curr->pi_state_list;
433 struct futex_pi_state *pi_state;
434 struct futex_hash_bucket *hb;
435 union futex_key key = FUTEX_KEY_INIT;
437 if (!futex_cmpxchg_enabled)
440 * We are a ZOMBIE and nobody can enqueue itself on
441 * pi_state_list anymore, but we have to be careful
442 * versus waiters unqueueing themselves:
444 spin_lock_irq(&curr->pi_lock);
445 while (!list_empty(head)) {
448 pi_state = list_entry(next, struct futex_pi_state, list);
450 hb = hash_futex(&key);
451 spin_unlock_irq(&curr->pi_lock);
453 spin_lock(&hb->lock);
455 spin_lock_irq(&curr->pi_lock);
457 * We dropped the pi-lock, so re-check whether this
458 * task still owns the PI-state:
460 if (head->next != next) {
461 spin_unlock(&hb->lock);
465 WARN_ON(pi_state->owner != curr);
466 WARN_ON(list_empty(&pi_state->list));
467 list_del_init(&pi_state->list);
468 pi_state->owner = NULL;
469 spin_unlock_irq(&curr->pi_lock);
471 rt_mutex_unlock(&pi_state->pi_mutex);
473 spin_unlock(&hb->lock);
475 spin_lock_irq(&curr->pi_lock);
477 spin_unlock_irq(&curr->pi_lock);
481 lookup_pi_state(u32 uval, struct futex_hash_bucket *hb,
482 union futex_key *key, struct futex_pi_state **ps)
484 struct futex_pi_state *pi_state = NULL;
485 struct futex_q *this, *next;
486 struct plist_head *head;
487 struct task_struct *p;
488 pid_t pid = uval & FUTEX_TID_MASK;
492 plist_for_each_entry_safe(this, next, head, list) {
493 if (match_futex(&this->key, key)) {
495 * Another waiter already exists - bump up
496 * the refcount and return its pi_state:
498 pi_state = this->pi_state;
500 * Userspace might have messed up non PI and PI futexes
502 if (unlikely(!pi_state))
505 WARN_ON(!atomic_read(&pi_state->refcount));
506 WARN_ON(pid && pi_state->owner &&
507 pi_state->owner->pid != pid);
509 atomic_inc(&pi_state->refcount);
517 * We are the first waiter - try to look up the real owner and attach
518 * the new pi_state to it, but bail out when TID = 0
522 p = futex_find_get_task(pid);
527 * We need to look at the task state flags to figure out,
528 * whether the task is exiting. To protect against the do_exit
529 * change of the task flags, we do this protected by
532 spin_lock_irq(&p->pi_lock);
533 if (unlikely(p->flags & PF_EXITING)) {
535 * The task is on the way out. When PF_EXITPIDONE is
536 * set, we know that the task has finished the
539 int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN;
541 spin_unlock_irq(&p->pi_lock);
546 pi_state = alloc_pi_state();
549 * Initialize the pi_mutex in locked state and make 'p'
552 rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p);
554 /* Store the key for possible exit cleanups: */
555 pi_state->key = *key;
557 WARN_ON(!list_empty(&pi_state->list));
558 list_add(&pi_state->list, &p->pi_state_list);
560 spin_unlock_irq(&p->pi_lock);
570 * The hash bucket lock must be held when this is called.
571 * Afterwards, the futex_q must not be accessed.
573 static void wake_futex(struct futex_q *q)
575 plist_del(&q->list, &q->list.plist);
577 * The lock in wake_up_all() is a crucial memory barrier after the
578 * plist_del() and also before assigning to q->lock_ptr.
582 * The waiting task can free the futex_q as soon as this is written,
583 * without taking any locks. This must come last.
585 * A memory barrier is required here to prevent the following store
586 * to lock_ptr from getting ahead of the wakeup. Clearing the lock
587 * at the end of wake_up_all() does not prevent this store from
594 static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this)
596 struct task_struct *new_owner;
597 struct futex_pi_state *pi_state = this->pi_state;
603 spin_lock(&pi_state->pi_mutex.wait_lock);
604 new_owner = rt_mutex_next_owner(&pi_state->pi_mutex);
607 * This happens when we have stolen the lock and the original
608 * pending owner did not enqueue itself back on the rt_mutex.
609 * Thats not a tragedy. We know that way, that a lock waiter
610 * is on the fly. We make the futex_q waiter the pending owner.
613 new_owner = this->task;
616 * We pass it to the next owner. (The WAITERS bit is always
617 * kept enabled while there is PI state around. We must also
618 * preserve the owner died bit.)
620 if (!(uval & FUTEX_OWNER_DIED)) {
623 newval = FUTEX_WAITERS | task_pid_vnr(new_owner);
625 curval = cmpxchg_futex_value_locked(uaddr, uval, newval);
627 if (curval == -EFAULT)
629 else if (curval != uval)
632 spin_unlock(&pi_state->pi_mutex.wait_lock);
637 spin_lock_irq(&pi_state->owner->pi_lock);
638 WARN_ON(list_empty(&pi_state->list));
639 list_del_init(&pi_state->list);
640 spin_unlock_irq(&pi_state->owner->pi_lock);
642 spin_lock_irq(&new_owner->pi_lock);
643 WARN_ON(!list_empty(&pi_state->list));
644 list_add(&pi_state->list, &new_owner->pi_state_list);
645 pi_state->owner = new_owner;
646 spin_unlock_irq(&new_owner->pi_lock);
648 spin_unlock(&pi_state->pi_mutex.wait_lock);
649 rt_mutex_unlock(&pi_state->pi_mutex);
654 static int unlock_futex_pi(u32 __user *uaddr, u32 uval)
659 * There is no waiter, so we unlock the futex. The owner died
660 * bit has not to be preserved here. We are the owner:
662 oldval = cmpxchg_futex_value_locked(uaddr, uval, 0);
664 if (oldval == -EFAULT)
673 * Express the locking dependencies for lockdep:
676 double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
679 spin_lock(&hb1->lock);
681 spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING);
682 } else { /* hb1 > hb2 */
683 spin_lock(&hb2->lock);
684 spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING);
689 * Wake up all waiters hashed on the physical page that is mapped
690 * to this virtual address:
692 static int futex_wake(u32 __user *uaddr, int fshared, int nr_wake, u32 bitset)
694 struct futex_hash_bucket *hb;
695 struct futex_q *this, *next;
696 struct plist_head *head;
697 union futex_key key = FUTEX_KEY_INIT;
703 ret = get_futex_key(uaddr, fshared, &key);
704 if (unlikely(ret != 0))
707 hb = hash_futex(&key);
708 spin_lock(&hb->lock);
711 plist_for_each_entry_safe(this, next, head, list) {
712 if (match_futex (&this->key, &key)) {
713 if (this->pi_state) {
718 /* Check if one of the bits is set in both bitsets */
719 if (!(this->bitset & bitset))
723 if (++ret >= nr_wake)
728 spin_unlock(&hb->lock);
729 put_futex_key(fshared, &key);
735 * Wake up all waiters hashed on the physical page that is mapped
736 * to this virtual address:
739 futex_wake_op(u32 __user *uaddr1, int fshared, u32 __user *uaddr2,
740 int nr_wake, int nr_wake2, int op)
742 union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
743 struct futex_hash_bucket *hb1, *hb2;
744 struct plist_head *head;
745 struct futex_q *this, *next;
746 int ret, op_ret, attempt = 0;
749 ret = get_futex_key(uaddr1, fshared, &key1);
750 if (unlikely(ret != 0))
752 ret = get_futex_key(uaddr2, fshared, &key2);
753 if (unlikely(ret != 0))
756 hb1 = hash_futex(&key1);
757 hb2 = hash_futex(&key2);
760 double_lock_hb(hb1, hb2);
762 op_ret = futex_atomic_op_inuser(op, uaddr2);
763 if (unlikely(op_ret < 0)) {
766 spin_unlock(&hb1->lock);
768 spin_unlock(&hb2->lock);
772 * we don't get EFAULT from MMU faults if we don't have an MMU,
773 * but we might get them from range checking
779 if (unlikely(op_ret != -EFAULT)) {
785 * futex_atomic_op_inuser needs to both read and write
786 * *(int __user *)uaddr2, but we can't modify it
787 * non-atomically. Therefore, if get_user below is not
788 * enough, we need to handle the fault ourselves, while
789 * still holding the mmap_sem.
792 ret = futex_handle_fault((unsigned long)uaddr2,
799 ret = get_user(dummy, uaddr2);
808 plist_for_each_entry_safe(this, next, head, list) {
809 if (match_futex (&this->key, &key1)) {
811 if (++ret >= nr_wake)
820 plist_for_each_entry_safe(this, next, head, list) {
821 if (match_futex (&this->key, &key2)) {
823 if (++op_ret >= nr_wake2)
830 spin_unlock(&hb1->lock);
832 spin_unlock(&hb2->lock);
834 put_futex_key(fshared, &key2);
836 put_futex_key(fshared, &key1);
842 * Requeue all waiters hashed on one physical page to another
845 static int futex_requeue(u32 __user *uaddr1, int fshared, u32 __user *uaddr2,
846 int nr_wake, int nr_requeue, u32 *cmpval)
848 union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
849 struct futex_hash_bucket *hb1, *hb2;
850 struct plist_head *head1;
851 struct futex_q *this, *next;
852 int ret, drop_count = 0;
855 ret = get_futex_key(uaddr1, fshared, &key1);
856 if (unlikely(ret != 0))
858 ret = get_futex_key(uaddr2, fshared, &key2);
859 if (unlikely(ret != 0))
862 hb1 = hash_futex(&key1);
863 hb2 = hash_futex(&key2);
865 double_lock_hb(hb1, hb2);
867 if (likely(cmpval != NULL)) {
870 ret = get_futex_value_locked(&curval, uaddr1);
873 spin_unlock(&hb1->lock);
875 spin_unlock(&hb2->lock);
877 ret = get_user(curval, uaddr1);
884 if (curval != *cmpval) {
891 plist_for_each_entry_safe(this, next, head1, list) {
892 if (!match_futex (&this->key, &key1))
894 if (++ret <= nr_wake) {
898 * If key1 and key2 hash to the same bucket, no need to
901 if (likely(head1 != &hb2->chain)) {
902 plist_del(&this->list, &hb1->chain);
903 plist_add(&this->list, &hb2->chain);
904 this->lock_ptr = &hb2->lock;
905 #ifdef CONFIG_DEBUG_PI_LIST
906 this->list.plist.lock = &hb2->lock;
910 get_futex_key_refs(&key2);
913 if (ret - nr_wake >= nr_requeue)
919 spin_unlock(&hb1->lock);
921 spin_unlock(&hb2->lock);
923 /* drop_futex_key_refs() must be called outside the spinlocks. */
924 while (--drop_count >= 0)
925 drop_futex_key_refs(&key1);
928 put_futex_key(fshared, &key2);
930 put_futex_key(fshared, &key1);
935 /* The key must be already stored in q->key. */
936 static inline struct futex_hash_bucket *queue_lock(struct futex_q *q)
938 struct futex_hash_bucket *hb;
940 init_waitqueue_head(&q->waiter);
942 get_futex_key_refs(&q->key);
943 hb = hash_futex(&q->key);
944 q->lock_ptr = &hb->lock;
946 spin_lock(&hb->lock);
950 static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb)
955 * The priority used to register this element is
956 * - either the real thread-priority for the real-time threads
957 * (i.e. threads with a priority lower than MAX_RT_PRIO)
958 * - or MAX_RT_PRIO for non-RT threads.
959 * Thus, all RT-threads are woken first in priority order, and
960 * the others are woken last, in FIFO order.
962 prio = min(current->normal_prio, MAX_RT_PRIO);
964 plist_node_init(&q->list, prio);
965 #ifdef CONFIG_DEBUG_PI_LIST
966 q->list.plist.lock = &hb->lock;
968 plist_add(&q->list, &hb->chain);
970 spin_unlock(&hb->lock);
974 queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb)
976 spin_unlock(&hb->lock);
977 drop_futex_key_refs(&q->key);
981 * queue_me and unqueue_me must be called as a pair, each
982 * exactly once. They are called with the hashed spinlock held.
985 /* Return 1 if we were still queued (ie. 0 means we were woken) */
986 static int unqueue_me(struct futex_q *q)
988 spinlock_t *lock_ptr;
991 /* In the common case we don't take the spinlock, which is nice. */
993 lock_ptr = q->lock_ptr;
995 if (lock_ptr != NULL) {
998 * q->lock_ptr can change between reading it and
999 * spin_lock(), causing us to take the wrong lock. This
1000 * corrects the race condition.
1002 * Reasoning goes like this: if we have the wrong lock,
1003 * q->lock_ptr must have changed (maybe several times)
1004 * between reading it and the spin_lock(). It can
1005 * change again after the spin_lock() but only if it was
1006 * already changed before the spin_lock(). It cannot,
1007 * however, change back to the original value. Therefore
1008 * we can detect whether we acquired the correct lock.
1010 if (unlikely(lock_ptr != q->lock_ptr)) {
1011 spin_unlock(lock_ptr);
1014 WARN_ON(plist_node_empty(&q->list));
1015 plist_del(&q->list, &q->list.plist);
1017 BUG_ON(q->pi_state);
1019 spin_unlock(lock_ptr);
1023 drop_futex_key_refs(&q->key);
1028 * PI futexes can not be requeued and must remove themself from the
1029 * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry
1032 static void unqueue_me_pi(struct futex_q *q)
1034 WARN_ON(plist_node_empty(&q->list));
1035 plist_del(&q->list, &q->list.plist);
1037 BUG_ON(!q->pi_state);
1038 free_pi_state(q->pi_state);
1041 spin_unlock(q->lock_ptr);
1043 drop_futex_key_refs(&q->key);
1047 * Fixup the pi_state owner with the new owner.
1049 * Must be called with hash bucket lock held and mm->sem held for non
1052 static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q,
1053 struct task_struct *newowner, int fshared)
1055 u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS;
1056 struct futex_pi_state *pi_state = q->pi_state;
1057 struct task_struct *oldowner = pi_state->owner;
1058 u32 uval, curval, newval;
1059 int ret, attempt = 0;
1062 if (!pi_state->owner)
1063 newtid |= FUTEX_OWNER_DIED;
1066 * We are here either because we stole the rtmutex from the
1067 * pending owner or we are the pending owner which failed to
1068 * get the rtmutex. We have to replace the pending owner TID
1069 * in the user space variable. This must be atomic as we have
1070 * to preserve the owner died bit here.
1072 * Note: We write the user space value _before_ changing the
1073 * pi_state because we can fault here. Imagine swapped out
1074 * pages or a fork, which was running right before we acquired
1075 * mmap_sem, that marked all the anonymous memory readonly for
1078 * Modifying pi_state _before_ the user space value would
1079 * leave the pi_state in an inconsistent state when we fault
1080 * here, because we need to drop the hash bucket lock to
1081 * handle the fault. This might be observed in the PID check
1082 * in lookup_pi_state.
1085 if (get_futex_value_locked(&uval, uaddr))
1089 newval = (uval & FUTEX_OWNER_DIED) | newtid;
1091 curval = cmpxchg_futex_value_locked(uaddr, uval, newval);
1093 if (curval == -EFAULT)
1101 * We fixed up user space. Now we need to fix the pi_state
1104 if (pi_state->owner != NULL) {
1105 spin_lock_irq(&pi_state->owner->pi_lock);
1106 WARN_ON(list_empty(&pi_state->list));
1107 list_del_init(&pi_state->list);
1108 spin_unlock_irq(&pi_state->owner->pi_lock);
1111 pi_state->owner = newowner;
1113 spin_lock_irq(&newowner->pi_lock);
1114 WARN_ON(!list_empty(&pi_state->list));
1115 list_add(&pi_state->list, &newowner->pi_state_list);
1116 spin_unlock_irq(&newowner->pi_lock);
1120 * To handle the page fault we need to drop the hash bucket
1121 * lock here. That gives the other task (either the pending
1122 * owner itself or the task which stole the rtmutex) the
1123 * chance to try the fixup of the pi_state. So once we are
1124 * back from handling the fault we need to check the pi_state
1125 * after reacquiring the hash bucket lock and before trying to
1126 * do another fixup. When the fixup has been done already we
1130 spin_unlock(q->lock_ptr);
1132 ret = futex_handle_fault((unsigned long)uaddr, attempt++);
1134 spin_lock(q->lock_ptr);
1137 * Check if someone else fixed it for us:
1139 if (pi_state->owner != oldowner)
1149 * In case we must use restart_block to restart a futex_wait,
1150 * we encode in the 'flags' shared capability
1152 #define FLAGS_SHARED 0x01
1153 #define FLAGS_CLOCKRT 0x02
1155 static long futex_wait_restart(struct restart_block *restart);
1157 static int futex_wait(u32 __user *uaddr, int fshared,
1158 u32 val, ktime_t *abs_time, u32 bitset, int clockrt)
1160 struct task_struct *curr = current;
1161 DECLARE_WAITQUEUE(wait, curr);
1162 struct futex_hash_bucket *hb;
1166 struct hrtimer_sleeper t;
1175 q.key = FUTEX_KEY_INIT;
1176 ret = get_futex_key(uaddr, fshared, &q.key);
1177 if (unlikely(ret != 0))
1180 hb = queue_lock(&q);
1183 * Access the page AFTER the futex is queued.
1184 * Order is important:
1186 * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val);
1187 * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); }
1189 * The basic logical guarantee of a futex is that it blocks ONLY
1190 * if cond(var) is known to be true at the time of blocking, for
1191 * any cond. If we queued after testing *uaddr, that would open
1192 * a race condition where we could block indefinitely with
1193 * cond(var) false, which would violate the guarantee.
1195 * A consequence is that futex_wait() can return zero and absorb
1196 * a wakeup when *uaddr != val on entry to the syscall. This is
1199 * for shared futexes, we hold the mmap semaphore, so the mapping
1200 * cannot have changed since we looked it up in get_futex_key.
1202 ret = get_futex_value_locked(&uval, uaddr);
1204 if (unlikely(ret)) {
1205 queue_unlock(&q, hb);
1206 put_futex_key(fshared, &q.key);
1208 ret = get_user(uval, uaddr);
1216 goto out_unlock_put_key;
1218 /* Only actually queue if *uaddr contained val. */
1222 * There might have been scheduling since the queue_me(), as we
1223 * cannot hold a spinlock across the get_user() in case it
1224 * faults, and we cannot just set TASK_INTERRUPTIBLE state when
1225 * queueing ourselves into the futex hash. This code thus has to
1226 * rely on the futex_wake() code removing us from hash when it
1230 /* add_wait_queue is the barrier after __set_current_state. */
1231 __set_current_state(TASK_INTERRUPTIBLE);
1232 add_wait_queue(&q.waiter, &wait);
1234 * !plist_node_empty() is safe here without any lock.
1235 * q.lock_ptr != 0 is not safe, because of ordering against wakeup.
1237 if (likely(!plist_node_empty(&q.list))) {
1241 unsigned long slack;
1242 slack = current->timer_slack_ns;
1243 if (rt_task(current))
1245 hrtimer_init_on_stack(&t.timer,
1246 clockrt ? CLOCK_REALTIME :
1249 hrtimer_init_sleeper(&t, current);
1250 hrtimer_set_expires_range_ns(&t.timer, *abs_time, slack);
1252 hrtimer_start_expires(&t.timer, HRTIMER_MODE_ABS);
1253 if (!hrtimer_active(&t.timer))
1257 * the timer could have already expired, in which
1258 * case current would be flagged for rescheduling.
1259 * Don't bother calling schedule.
1264 hrtimer_cancel(&t.timer);
1266 /* Flag if a timeout occured */
1267 rem = (t.task == NULL);
1269 destroy_hrtimer_on_stack(&t.timer);
1272 __set_current_state(TASK_RUNNING);
1275 * NOTE: we don't remove ourselves from the waitqueue because
1276 * we are the only user of it.
1279 /* If we were woken (and unqueued), we succeeded, whatever. */
1280 if (!unqueue_me(&q))
1286 * We expect signal_pending(current), but another thread may
1287 * have handled it for us already.
1290 return -ERESTARTSYS;
1292 struct restart_block *restart;
1293 restart = ¤t_thread_info()->restart_block;
1294 restart->fn = futex_wait_restart;
1295 restart->futex.uaddr = (u32 *)uaddr;
1296 restart->futex.val = val;
1297 restart->futex.time = abs_time->tv64;
1298 restart->futex.bitset = bitset;
1299 restart->futex.flags = 0;
1302 restart->futex.flags |= FLAGS_SHARED;
1304 restart->futex.flags |= FLAGS_CLOCKRT;
1305 return -ERESTART_RESTARTBLOCK;
1309 queue_unlock(&q, hb);
1310 put_futex_key(fshared, &q.key);
1317 static long futex_wait_restart(struct restart_block *restart)
1319 u32 __user *uaddr = (u32 __user *)restart->futex.uaddr;
1323 t.tv64 = restart->futex.time;
1324 restart->fn = do_no_restart_syscall;
1325 if (restart->futex.flags & FLAGS_SHARED)
1327 return (long)futex_wait(uaddr, fshared, restart->futex.val, &t,
1328 restart->futex.bitset,
1329 restart->futex.flags & FLAGS_CLOCKRT);
1334 * Userspace tried a 0 -> TID atomic transition of the futex value
1335 * and failed. The kernel side here does the whole locking operation:
1336 * if there are waiters then it will block, it does PI, etc. (Due to
1337 * races the kernel might see a 0 value of the futex too.)
1339 static int futex_lock_pi(u32 __user *uaddr, int fshared,
1340 int detect, ktime_t *time, int trylock)
1342 struct hrtimer_sleeper timeout, *to = NULL;
1343 struct task_struct *curr = current;
1344 struct futex_hash_bucket *hb;
1345 u32 uval, newval, curval;
1347 int ret, lock_taken, ownerdied = 0, attempt = 0;
1349 if (refill_pi_state_cache())
1354 hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME,
1356 hrtimer_init_sleeper(to, current);
1357 hrtimer_set_expires(&to->timer, *time);
1362 q.key = FUTEX_KEY_INIT;
1363 ret = get_futex_key(uaddr, fshared, &q.key);
1364 if (unlikely(ret != 0))
1368 hb = queue_lock(&q);
1371 ret = lock_taken = 0;
1374 * To avoid races, we attempt to take the lock here again
1375 * (by doing a 0 -> TID atomic cmpxchg), while holding all
1376 * the locks. It will most likely not succeed.
1378 newval = task_pid_vnr(current);
1380 curval = cmpxchg_futex_value_locked(uaddr, 0, newval);
1382 if (unlikely(curval == -EFAULT))
1386 * Detect deadlocks. In case of REQUEUE_PI this is a valid
1387 * situation and we return success to user space.
1389 if (unlikely((curval & FUTEX_TID_MASK) == task_pid_vnr(current))) {
1391 goto out_unlock_put_key;
1395 * Surprise - we got the lock. Just return to userspace:
1397 if (unlikely(!curval))
1398 goto out_unlock_put_key;
1403 * Set the WAITERS flag, so the owner will know it has someone
1404 * to wake at next unlock
1406 newval = curval | FUTEX_WAITERS;
1409 * There are two cases, where a futex might have no owner (the
1410 * owner TID is 0): OWNER_DIED. We take over the futex in this
1411 * case. We also do an unconditional take over, when the owner
1412 * of the futex died.
1414 * This is safe as we are protected by the hash bucket lock !
1416 if (unlikely(ownerdied || !(curval & FUTEX_TID_MASK))) {
1417 /* Keep the OWNER_DIED bit */
1418 newval = (curval & ~FUTEX_TID_MASK) | task_pid_vnr(current);
1423 curval = cmpxchg_futex_value_locked(uaddr, uval, newval);
1425 if (unlikely(curval == -EFAULT))
1427 if (unlikely(curval != uval))
1431 * We took the lock due to owner died take over.
1433 if (unlikely(lock_taken))
1434 goto out_unlock_put_key;
1437 * We dont have the lock. Look up the PI state (or create it if
1438 * we are the first waiter):
1440 ret = lookup_pi_state(uval, hb, &q.key, &q.pi_state);
1442 if (unlikely(ret)) {
1447 * Task is exiting and we just wait for the
1450 queue_unlock(&q, hb);
1456 * No owner found for this futex. Check if the
1457 * OWNER_DIED bit is set to figure out whether
1458 * this is a robust futex or not.
1460 if (get_futex_value_locked(&curval, uaddr))
1464 * We simply start over in case of a robust
1465 * futex. The code above will take the futex
1468 if (curval & FUTEX_OWNER_DIED) {
1473 goto out_unlock_put_key;
1478 * Only actually queue now that the atomic ops are done:
1482 WARN_ON(!q.pi_state);
1484 * Block on the PI mutex:
1487 ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1);
1489 ret = rt_mutex_trylock(&q.pi_state->pi_mutex);
1490 /* Fixup the trylock return value: */
1491 ret = ret ? 0 : -EWOULDBLOCK;
1494 spin_lock(q.lock_ptr);
1498 * Got the lock. We might not be the anticipated owner
1499 * if we did a lock-steal - fix up the PI-state in
1502 if (q.pi_state->owner != curr)
1503 ret = fixup_pi_state_owner(uaddr, &q, curr, fshared);
1506 * Catch the rare case, where the lock was released
1507 * when we were on the way back before we locked the
1510 if (q.pi_state->owner == curr) {
1512 * Try to get the rt_mutex now. This might
1513 * fail as some other task acquired the
1514 * rt_mutex after we removed ourself from the
1515 * rt_mutex waiters list.
1517 if (rt_mutex_trylock(&q.pi_state->pi_mutex))
1521 * pi_state is incorrect, some other
1522 * task did a lock steal and we
1523 * returned due to timeout or signal
1524 * without taking the rt_mutex. Too
1525 * late. We can access the
1526 * rt_mutex_owner without locking, as
1527 * the other task is now blocked on
1528 * the hash bucket lock. Fix the state
1531 struct task_struct *owner;
1534 owner = rt_mutex_owner(&q.pi_state->pi_mutex);
1535 res = fixup_pi_state_owner(uaddr, &q, owner,
1538 /* propagate -EFAULT, if the fixup failed */
1544 * Paranoia check. If we did not take the lock
1545 * in the trylock above, then we should not be
1546 * the owner of the rtmutex, neither the real
1547 * nor the pending one:
1549 if (rt_mutex_owner(&q.pi_state->pi_mutex) == curr)
1550 printk(KERN_ERR "futex_lock_pi: ret = %d "
1551 "pi-mutex: %p pi-state %p\n", ret,
1552 q.pi_state->pi_mutex.owner,
1557 /* Unqueue and drop the lock */
1561 destroy_hrtimer_on_stack(&to->timer);
1562 return ret != -EINTR ? ret : -ERESTARTNOINTR;
1565 queue_unlock(&q, hb);
1568 put_futex_key(fshared, &q.key);
1571 destroy_hrtimer_on_stack(&to->timer);
1576 * We have to r/w *(int __user *)uaddr, and we have to modify it
1577 * atomically. Therefore, if we continue to fault after get_user()
1578 * below, we need to handle the fault ourselves, while still holding
1579 * the mmap_sem. This can occur if the uaddr is under contention as
1580 * we have to drop the mmap_sem in order to call get_user().
1582 queue_unlock(&q, hb);
1585 ret = futex_handle_fault((unsigned long)uaddr, attempt);
1588 goto retry_unlocked;
1591 ret = get_user(uval, uaddr);
1596 destroy_hrtimer_on_stack(&to->timer);
1601 * Userspace attempted a TID -> 0 atomic transition, and failed.
1602 * This is the in-kernel slowpath: we look up the PI state (if any),
1603 * and do the rt-mutex unlock.
1605 static int futex_unlock_pi(u32 __user *uaddr, int fshared)
1607 struct futex_hash_bucket *hb;
1608 struct futex_q *this, *next;
1610 struct plist_head *head;
1611 union futex_key key = FUTEX_KEY_INIT;
1612 int ret, attempt = 0;
1615 if (get_user(uval, uaddr))
1618 * We release only a lock we actually own:
1620 if ((uval & FUTEX_TID_MASK) != task_pid_vnr(current))
1623 ret = get_futex_key(uaddr, fshared, &key);
1624 if (unlikely(ret != 0))
1627 hb = hash_futex(&key);
1629 spin_lock(&hb->lock);
1632 * To avoid races, try to do the TID -> 0 atomic transition
1633 * again. If it succeeds then we can return without waking
1636 if (!(uval & FUTEX_OWNER_DIED))
1637 uval = cmpxchg_futex_value_locked(uaddr, task_pid_vnr(current), 0);
1640 if (unlikely(uval == -EFAULT))
1643 * Rare case: we managed to release the lock atomically,
1644 * no need to wake anyone else up:
1646 if (unlikely(uval == task_pid_vnr(current)))
1650 * Ok, other tasks may need to be woken up - check waiters
1651 * and do the wakeup if necessary:
1655 plist_for_each_entry_safe(this, next, head, list) {
1656 if (!match_futex (&this->key, &key))
1658 ret = wake_futex_pi(uaddr, uval, this);
1660 * The atomic access to the futex value
1661 * generated a pagefault, so retry the
1662 * user-access and the wakeup:
1669 * No waiters - kernel unlocks the futex:
1671 if (!(uval & FUTEX_OWNER_DIED)) {
1672 ret = unlock_futex_pi(uaddr, uval);
1678 spin_unlock(&hb->lock);
1679 put_futex_key(fshared, &key);
1686 * We have to r/w *(int __user *)uaddr, and we have to modify it
1687 * atomically. Therefore, if we continue to fault after get_user()
1688 * below, we need to handle the fault ourselves, while still holding
1689 * the mmap_sem. This can occur if the uaddr is under contention as
1690 * we have to drop the mmap_sem in order to call get_user().
1692 spin_unlock(&hb->lock);
1695 ret = futex_handle_fault((unsigned long)uaddr, attempt);
1699 goto retry_unlocked;
1702 ret = get_user(uval, uaddr);
1710 * Support for robust futexes: the kernel cleans up held futexes at
1713 * Implementation: user-space maintains a per-thread list of locks it
1714 * is holding. Upon do_exit(), the kernel carefully walks this list,
1715 * and marks all locks that are owned by this thread with the
1716 * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is
1717 * always manipulated with the lock held, so the list is private and
1718 * per-thread. Userspace also maintains a per-thread 'list_op_pending'
1719 * field, to allow the kernel to clean up if the thread dies after
1720 * acquiring the lock, but just before it could have added itself to
1721 * the list. There can only be one such pending lock.
1725 * sys_set_robust_list - set the robust-futex list head of a task
1726 * @head: pointer to the list-head
1727 * @len: length of the list-head, as userspace expects
1730 sys_set_robust_list(struct robust_list_head __user *head,
1733 if (!futex_cmpxchg_enabled)
1736 * The kernel knows only one size for now:
1738 if (unlikely(len != sizeof(*head)))
1741 current->robust_list = head;
1747 * sys_get_robust_list - get the robust-futex list head of a task
1748 * @pid: pid of the process [zero for current task]
1749 * @head_ptr: pointer to a list-head pointer, the kernel fills it in
1750 * @len_ptr: pointer to a length field, the kernel fills in the header size
1753 sys_get_robust_list(int pid, struct robust_list_head __user * __user *head_ptr,
1754 size_t __user *len_ptr)
1756 struct robust_list_head __user *head;
1759 if (!futex_cmpxchg_enabled)
1763 head = current->robust_list;
1765 struct task_struct *p;
1769 p = find_task_by_vpid(pid);
1773 if ((current->euid != p->euid) && (current->euid != p->uid) &&
1774 !capable(CAP_SYS_PTRACE))
1776 head = p->robust_list;
1780 if (put_user(sizeof(*head), len_ptr))
1782 return put_user(head, head_ptr);
1791 * Process a futex-list entry, check whether it's owned by the
1792 * dying task, and do notification if so:
1794 int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi)
1796 u32 uval, nval, mval;
1799 if (get_user(uval, uaddr))
1802 if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) {
1804 * Ok, this dying thread is truly holding a futex
1805 * of interest. Set the OWNER_DIED bit atomically
1806 * via cmpxchg, and if the value had FUTEX_WAITERS
1807 * set, wake up a waiter (if any). (We have to do a
1808 * futex_wake() even if OWNER_DIED is already set -
1809 * to handle the rare but possible case of recursive
1810 * thread-death.) The rest of the cleanup is done in
1813 mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED;
1814 nval = futex_atomic_cmpxchg_inatomic(uaddr, uval, mval);
1816 if (nval == -EFAULT)
1823 * Wake robust non-PI futexes here. The wakeup of
1824 * PI futexes happens in exit_pi_state():
1826 if (!pi && (uval & FUTEX_WAITERS))
1827 futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY);
1833 * Fetch a robust-list pointer. Bit 0 signals PI futexes:
1835 static inline int fetch_robust_entry(struct robust_list __user **entry,
1836 struct robust_list __user * __user *head,
1839 unsigned long uentry;
1841 if (get_user(uentry, (unsigned long __user *)head))
1844 *entry = (void __user *)(uentry & ~1UL);
1851 * Walk curr->robust_list (very carefully, it's a userspace list!)
1852 * and mark any locks found there dead, and notify any waiters.
1854 * We silently return on any sign of list-walking problem.
1856 void exit_robust_list(struct task_struct *curr)
1858 struct robust_list_head __user *head = curr->robust_list;
1859 struct robust_list __user *entry, *next_entry, *pending;
1860 unsigned int limit = ROBUST_LIST_LIMIT, pi, next_pi, pip;
1861 unsigned long futex_offset;
1864 if (!futex_cmpxchg_enabled)
1868 * Fetch the list head (which was registered earlier, via
1869 * sys_set_robust_list()):
1871 if (fetch_robust_entry(&entry, &head->list.next, &pi))
1874 * Fetch the relative futex offset:
1876 if (get_user(futex_offset, &head->futex_offset))
1879 * Fetch any possibly pending lock-add first, and handle it
1882 if (fetch_robust_entry(&pending, &head->list_op_pending, &pip))
1885 next_entry = NULL; /* avoid warning with gcc */
1886 while (entry != &head->list) {
1888 * Fetch the next entry in the list before calling
1889 * handle_futex_death:
1891 rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi);
1893 * A pending lock might already be on the list, so
1894 * don't process it twice:
1896 if (entry != pending)
1897 if (handle_futex_death((void __user *)entry + futex_offset,
1905 * Avoid excessively long or circular lists:
1914 handle_futex_death((void __user *)pending + futex_offset,
1918 long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout,
1919 u32 __user *uaddr2, u32 val2, u32 val3)
1921 int clockrt, ret = -ENOSYS;
1922 int cmd = op & FUTEX_CMD_MASK;
1925 if (!(op & FUTEX_PRIVATE_FLAG))
1928 clockrt = op & FUTEX_CLOCK_REALTIME;
1929 if (clockrt && cmd != FUTEX_WAIT_BITSET)
1934 val3 = FUTEX_BITSET_MATCH_ANY;
1935 case FUTEX_WAIT_BITSET:
1936 ret = futex_wait(uaddr, fshared, val, timeout, val3, clockrt);
1939 val3 = FUTEX_BITSET_MATCH_ANY;
1940 case FUTEX_WAKE_BITSET:
1941 ret = futex_wake(uaddr, fshared, val, val3);
1944 ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, NULL);
1946 case FUTEX_CMP_REQUEUE:
1947 ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, &val3);
1950 ret = futex_wake_op(uaddr, fshared, uaddr2, val, val2, val3);
1953 if (futex_cmpxchg_enabled)
1954 ret = futex_lock_pi(uaddr, fshared, val, timeout, 0);
1956 case FUTEX_UNLOCK_PI:
1957 if (futex_cmpxchg_enabled)
1958 ret = futex_unlock_pi(uaddr, fshared);
1960 case FUTEX_TRYLOCK_PI:
1961 if (futex_cmpxchg_enabled)
1962 ret = futex_lock_pi(uaddr, fshared, 0, timeout, 1);
1971 asmlinkage long sys_futex(u32 __user *uaddr, int op, u32 val,
1972 struct timespec __user *utime, u32 __user *uaddr2,
1976 ktime_t t, *tp = NULL;
1978 int cmd = op & FUTEX_CMD_MASK;
1980 if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI ||
1981 cmd == FUTEX_WAIT_BITSET)) {
1982 if (copy_from_user(&ts, utime, sizeof(ts)) != 0)
1984 if (!timespec_valid(&ts))
1987 t = timespec_to_ktime(ts);
1988 if (cmd == FUTEX_WAIT)
1989 t = ktime_add_safe(ktime_get(), t);
1993 * requeue parameter in 'utime' if cmd == FUTEX_REQUEUE.
1994 * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP.
1996 if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE ||
1997 cmd == FUTEX_WAKE_OP)
1998 val2 = (u32) (unsigned long) utime;
2000 return do_futex(uaddr, op, val, tp, uaddr2, val2, val3);
2003 static int __init futex_init(void)
2009 * This will fail and we want it. Some arch implementations do
2010 * runtime detection of the futex_atomic_cmpxchg_inatomic()
2011 * functionality. We want to know that before we call in any
2012 * of the complex code paths. Also we want to prevent
2013 * registration of robust lists in that case. NULL is
2014 * guaranteed to fault and we get -EFAULT on functional
2015 * implementation, the non functional ones will return
2018 curval = cmpxchg_futex_value_locked(NULL, 0, 0);
2019 if (curval == -EFAULT)
2020 futex_cmpxchg_enabled = 1;
2022 for (i = 0; i < ARRAY_SIZE(futex_queues); i++) {
2023 plist_head_init(&futex_queues[i].chain, &futex_queues[i].lock);
2024 spin_lock_init(&futex_queues[i].lock);
2029 __initcall(futex_init);