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->waiters, then make the second condition true.
98 struct plist_node list;
99 wait_queue_head_t waiters;
101 /* Which hash list lock to use: */
102 spinlock_t *lock_ptr;
104 /* Key which the futex is hashed on: */
107 /* Optional priority inheritance state: */
108 struct futex_pi_state *pi_state;
109 struct task_struct *task;
111 /* Bitset for the optional bitmasked wakeup */
116 * Split the global futex_lock into every hash list lock.
118 struct futex_hash_bucket {
120 struct plist_head chain;
123 static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS];
126 * We hash on the keys returned from get_futex_key (see below).
128 static struct futex_hash_bucket *hash_futex(union futex_key *key)
130 u32 hash = jhash2((u32*)&key->both.word,
131 (sizeof(key->both.word)+sizeof(key->both.ptr))/4,
133 return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)];
137 * Return 1 if two futex_keys are equal, 0 otherwise.
139 static inline int match_futex(union futex_key *key1, union futex_key *key2)
141 return (key1->both.word == key2->both.word
142 && key1->both.ptr == key2->both.ptr
143 && key1->both.offset == key2->both.offset);
147 * Take a reference to the resource addressed by a key.
148 * Can be called while holding spinlocks.
151 static void get_futex_key_refs(union futex_key *key)
156 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
158 atomic_inc(&key->shared.inode->i_count);
160 case FUT_OFF_MMSHARED:
161 atomic_inc(&key->private.mm->mm_count);
167 * Drop a reference to the resource addressed by a key.
168 * The hash bucket spinlock must not be held.
170 static void drop_futex_key_refs(union futex_key *key)
175 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
177 iput(key->shared.inode);
179 case FUT_OFF_MMSHARED:
180 mmdrop(key->private.mm);
186 * get_futex_key - Get parameters which are the keys for a futex.
187 * @uaddr: virtual address of the futex
188 * @shared: NULL for a PROCESS_PRIVATE futex,
189 * ¤t->mm->mmap_sem for a PROCESS_SHARED futex
190 * @key: address where result is stored.
192 * Returns a negative error code or 0
193 * The key words are stored in *key on success.
195 * For shared mappings, it's (page->index, vma->vm_file->f_path.dentry->d_inode,
196 * offset_within_page). For private mappings, it's (uaddr, current->mm).
197 * We can usually work out the index without swapping in the page.
199 * fshared is NULL for PROCESS_PRIVATE futexes
200 * For other futexes, it points to ¤t->mm->mmap_sem and
201 * caller must have taken the reader lock. but NOT any spinlocks.
203 static int get_futex_key(u32 __user *uaddr, struct rw_semaphore *fshared,
204 union futex_key *key)
206 unsigned long address = (unsigned long)uaddr;
207 struct mm_struct *mm = current->mm;
212 * The futex address must be "naturally" aligned.
214 key->both.offset = address % PAGE_SIZE;
215 if (unlikely((address % sizeof(u32)) != 0))
217 address -= key->both.offset;
220 * PROCESS_PRIVATE futexes are fast.
221 * As the mm cannot disappear under us and the 'key' only needs
222 * virtual address, we dont even have to find the underlying vma.
223 * Note : We do have to check 'uaddr' is a valid user address,
224 * but access_ok() should be faster than find_vma()
227 if (unlikely(!access_ok(VERIFY_WRITE, uaddr, sizeof(u32))))
229 key->private.mm = mm;
230 key->private.address = address;
235 err = get_user_pages_fast(address, 1, 0, &page);
240 if (!page->mapping) {
247 * Private mappings are handled in a simple way.
249 * NOTE: When userspace waits on a MAP_SHARED mapping, even if
250 * it's a read-only handle, it's expected that futexes attach to
251 * the object not the particular process.
253 if (PageAnon(page)) {
254 key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */
255 key->private.mm = mm;
256 key->private.address = address;
258 key->both.offset |= FUT_OFF_INODE; /* inode-based key */
259 key->shared.inode = page->mapping->host;
260 key->shared.pgoff = page->index;
263 get_futex_key_refs(key);
271 void put_futex_key(struct rw_semaphore *fshared, union futex_key *key)
273 drop_futex_key_refs(key);
276 static u32 cmpxchg_futex_value_locked(u32 __user *uaddr, u32 uval, u32 newval)
281 curval = futex_atomic_cmpxchg_inatomic(uaddr, uval, newval);
287 static int get_futex_value_locked(u32 *dest, u32 __user *from)
292 ret = __copy_from_user_inatomic(dest, from, sizeof(u32));
295 return ret ? -EFAULT : 0;
300 * if fshared is non NULL, current->mm->mmap_sem is already held
302 static int futex_handle_fault(unsigned long address,
303 struct rw_semaphore *fshared, int attempt)
305 struct vm_area_struct * vma;
306 struct mm_struct *mm = current->mm;
312 down_read(&mm->mmap_sem);
313 vma = find_vma(mm, address);
314 if (vma && address >= vma->vm_start &&
315 (vma->vm_flags & VM_WRITE)) {
317 fault = handle_mm_fault(mm, vma, address, 1);
318 if (unlikely((fault & VM_FAULT_ERROR))) {
320 /* XXX: let's do this when we verify it is OK */
321 if (ret & VM_FAULT_OOM)
326 if (fault & VM_FAULT_MAJOR)
332 up_read(&mm->mmap_sem);
339 static int refill_pi_state_cache(void)
341 struct futex_pi_state *pi_state;
343 if (likely(current->pi_state_cache))
346 pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL);
351 INIT_LIST_HEAD(&pi_state->list);
352 /* pi_mutex gets initialized later */
353 pi_state->owner = NULL;
354 atomic_set(&pi_state->refcount, 1);
355 pi_state->key = FUTEX_KEY_INIT;
357 current->pi_state_cache = pi_state;
362 static struct futex_pi_state * alloc_pi_state(void)
364 struct futex_pi_state *pi_state = current->pi_state_cache;
367 current->pi_state_cache = NULL;
372 static void free_pi_state(struct futex_pi_state *pi_state)
374 if (!atomic_dec_and_test(&pi_state->refcount))
378 * If pi_state->owner is NULL, the owner is most probably dying
379 * and has cleaned up the pi_state already
381 if (pi_state->owner) {
382 spin_lock_irq(&pi_state->owner->pi_lock);
383 list_del_init(&pi_state->list);
384 spin_unlock_irq(&pi_state->owner->pi_lock);
386 rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner);
389 if (current->pi_state_cache)
393 * pi_state->list is already empty.
394 * clear pi_state->owner.
395 * refcount is at 0 - put it back to 1.
397 pi_state->owner = NULL;
398 atomic_set(&pi_state->refcount, 1);
399 current->pi_state_cache = pi_state;
404 * Look up the task based on what TID userspace gave us.
407 static struct task_struct * futex_find_get_task(pid_t pid)
409 struct task_struct *p;
412 p = find_task_by_vpid(pid);
413 if (!p || ((current->euid != p->euid) && (current->euid != p->uid)))
424 * This task is holding PI mutexes at exit time => bad.
425 * Kernel cleans up PI-state, but userspace is likely hosed.
426 * (Robust-futex cleanup is separate and might save the day for userspace.)
428 void exit_pi_state_list(struct task_struct *curr)
430 struct list_head *next, *head = &curr->pi_state_list;
431 struct futex_pi_state *pi_state;
432 struct futex_hash_bucket *hb;
433 union futex_key key = FUTEX_KEY_INIT;
435 if (!futex_cmpxchg_enabled)
438 * We are a ZOMBIE and nobody can enqueue itself on
439 * pi_state_list anymore, but we have to be careful
440 * versus waiters unqueueing themselves:
442 spin_lock_irq(&curr->pi_lock);
443 while (!list_empty(head)) {
446 pi_state = list_entry(next, struct futex_pi_state, list);
448 hb = hash_futex(&key);
449 spin_unlock_irq(&curr->pi_lock);
451 spin_lock(&hb->lock);
453 spin_lock_irq(&curr->pi_lock);
455 * We dropped the pi-lock, so re-check whether this
456 * task still owns the PI-state:
458 if (head->next != next) {
459 spin_unlock(&hb->lock);
463 WARN_ON(pi_state->owner != curr);
464 WARN_ON(list_empty(&pi_state->list));
465 list_del_init(&pi_state->list);
466 pi_state->owner = NULL;
467 spin_unlock_irq(&curr->pi_lock);
469 rt_mutex_unlock(&pi_state->pi_mutex);
471 spin_unlock(&hb->lock);
473 spin_lock_irq(&curr->pi_lock);
475 spin_unlock_irq(&curr->pi_lock);
479 lookup_pi_state(u32 uval, struct futex_hash_bucket *hb,
480 union futex_key *key, struct futex_pi_state **ps)
482 struct futex_pi_state *pi_state = NULL;
483 struct futex_q *this, *next;
484 struct plist_head *head;
485 struct task_struct *p;
486 pid_t pid = uval & FUTEX_TID_MASK;
490 plist_for_each_entry_safe(this, next, head, list) {
491 if (match_futex(&this->key, key)) {
493 * Another waiter already exists - bump up
494 * the refcount and return its pi_state:
496 pi_state = this->pi_state;
498 * Userspace might have messed up non PI and PI futexes
500 if (unlikely(!pi_state))
503 WARN_ON(!atomic_read(&pi_state->refcount));
504 WARN_ON(pid && pi_state->owner &&
505 pi_state->owner->pid != pid);
507 atomic_inc(&pi_state->refcount);
515 * We are the first waiter - try to look up the real owner and attach
516 * the new pi_state to it, but bail out when TID = 0
520 p = futex_find_get_task(pid);
525 * We need to look at the task state flags to figure out,
526 * whether the task is exiting. To protect against the do_exit
527 * change of the task flags, we do this protected by
530 spin_lock_irq(&p->pi_lock);
531 if (unlikely(p->flags & PF_EXITING)) {
533 * The task is on the way out. When PF_EXITPIDONE is
534 * set, we know that the task has finished the
537 int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN;
539 spin_unlock_irq(&p->pi_lock);
544 pi_state = alloc_pi_state();
547 * Initialize the pi_mutex in locked state and make 'p'
550 rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p);
552 /* Store the key for possible exit cleanups: */
553 pi_state->key = *key;
555 WARN_ON(!list_empty(&pi_state->list));
556 list_add(&pi_state->list, &p->pi_state_list);
558 spin_unlock_irq(&p->pi_lock);
568 * The hash bucket lock must be held when this is called.
569 * Afterwards, the futex_q must not be accessed.
571 static void wake_futex(struct futex_q *q)
573 plist_del(&q->list, &q->list.plist);
575 * The lock in wake_up_all() is a crucial memory barrier after the
576 * plist_del() and also before assigning to q->lock_ptr.
578 wake_up_all(&q->waiters);
580 * The waiting task can free the futex_q as soon as this is written,
581 * without taking any locks. This must come last.
583 * A memory barrier is required here to prevent the following store
584 * to lock_ptr from getting ahead of the wakeup. Clearing the lock
585 * at the end of wake_up_all() does not prevent this store from
592 static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this)
594 struct task_struct *new_owner;
595 struct futex_pi_state *pi_state = this->pi_state;
601 spin_lock(&pi_state->pi_mutex.wait_lock);
602 new_owner = rt_mutex_next_owner(&pi_state->pi_mutex);
605 * This happens when we have stolen the lock and the original
606 * pending owner did not enqueue itself back on the rt_mutex.
607 * Thats not a tragedy. We know that way, that a lock waiter
608 * is on the fly. We make the futex_q waiter the pending owner.
611 new_owner = this->task;
614 * We pass it to the next owner. (The WAITERS bit is always
615 * kept enabled while there is PI state around. We must also
616 * preserve the owner died bit.)
618 if (!(uval & FUTEX_OWNER_DIED)) {
621 newval = FUTEX_WAITERS | task_pid_vnr(new_owner);
623 curval = cmpxchg_futex_value_locked(uaddr, uval, newval);
625 if (curval == -EFAULT)
627 else if (curval != uval)
630 spin_unlock(&pi_state->pi_mutex.wait_lock);
635 spin_lock_irq(&pi_state->owner->pi_lock);
636 WARN_ON(list_empty(&pi_state->list));
637 list_del_init(&pi_state->list);
638 spin_unlock_irq(&pi_state->owner->pi_lock);
640 spin_lock_irq(&new_owner->pi_lock);
641 WARN_ON(!list_empty(&pi_state->list));
642 list_add(&pi_state->list, &new_owner->pi_state_list);
643 pi_state->owner = new_owner;
644 spin_unlock_irq(&new_owner->pi_lock);
646 spin_unlock(&pi_state->pi_mutex.wait_lock);
647 rt_mutex_unlock(&pi_state->pi_mutex);
652 static int unlock_futex_pi(u32 __user *uaddr, u32 uval)
657 * There is no waiter, so we unlock the futex. The owner died
658 * bit has not to be preserved here. We are the owner:
660 oldval = cmpxchg_futex_value_locked(uaddr, uval, 0);
662 if (oldval == -EFAULT)
671 * Express the locking dependencies for lockdep:
674 double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
677 spin_lock(&hb1->lock);
679 spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING);
680 } else { /* hb1 > hb2 */
681 spin_lock(&hb2->lock);
682 spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING);
687 * Wake up all waiters hashed on the physical page that is mapped
688 * to this virtual address:
690 static int futex_wake(u32 __user *uaddr, struct rw_semaphore *fshared,
691 int nr_wake, u32 bitset)
693 struct futex_hash_bucket *hb;
694 struct futex_q *this, *next;
695 struct plist_head *head;
696 union futex_key key = FUTEX_KEY_INIT;
702 ret = get_futex_key(uaddr, fshared, &key);
703 if (unlikely(ret != 0))
706 hb = hash_futex(&key);
707 spin_lock(&hb->lock);
710 plist_for_each_entry_safe(this, next, head, list) {
711 if (match_futex (&this->key, &key)) {
712 if (this->pi_state) {
717 /* Check if one of the bits is set in both bitsets */
718 if (!(this->bitset & bitset))
722 if (++ret >= nr_wake)
727 spin_unlock(&hb->lock);
729 put_futex_key(fshared, &key);
734 * Wake up all waiters hashed on the physical page that is mapped
735 * to this virtual address:
738 futex_wake_op(u32 __user *uaddr1, struct rw_semaphore *fshared,
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);
835 put_futex_key(fshared, &key1);
841 * Requeue all waiters hashed on one physical page to another
844 static int futex_requeue(u32 __user *uaddr1, struct rw_semaphore *fshared,
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);
929 put_futex_key(fshared, &key1);
933 /* The key must be already stored in q->key. */
934 static inline struct futex_hash_bucket *queue_lock(struct futex_q *q)
936 struct futex_hash_bucket *hb;
938 init_waitqueue_head(&q->waiters);
940 get_futex_key_refs(&q->key);
941 hb = hash_futex(&q->key);
942 q->lock_ptr = &hb->lock;
944 spin_lock(&hb->lock);
948 static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb)
953 * The priority used to register this element is
954 * - either the real thread-priority for the real-time threads
955 * (i.e. threads with a priority lower than MAX_RT_PRIO)
956 * - or MAX_RT_PRIO for non-RT threads.
957 * Thus, all RT-threads are woken first in priority order, and
958 * the others are woken last, in FIFO order.
960 prio = min(current->normal_prio, MAX_RT_PRIO);
962 plist_node_init(&q->list, prio);
963 #ifdef CONFIG_DEBUG_PI_LIST
964 q->list.plist.lock = &hb->lock;
966 plist_add(&q->list, &hb->chain);
968 spin_unlock(&hb->lock);
972 queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb)
974 spin_unlock(&hb->lock);
975 drop_futex_key_refs(&q->key);
979 * queue_me and unqueue_me must be called as a pair, each
980 * exactly once. They are called with the hashed spinlock held.
983 /* Return 1 if we were still queued (ie. 0 means we were woken) */
984 static int unqueue_me(struct futex_q *q)
986 spinlock_t *lock_ptr;
989 /* In the common case we don't take the spinlock, which is nice. */
991 lock_ptr = q->lock_ptr;
993 if (lock_ptr != NULL) {
996 * q->lock_ptr can change between reading it and
997 * spin_lock(), causing us to take the wrong lock. This
998 * corrects the race condition.
1000 * Reasoning goes like this: if we have the wrong lock,
1001 * q->lock_ptr must have changed (maybe several times)
1002 * between reading it and the spin_lock(). It can
1003 * change again after the spin_lock() but only if it was
1004 * already changed before the spin_lock(). It cannot,
1005 * however, change back to the original value. Therefore
1006 * we can detect whether we acquired the correct lock.
1008 if (unlikely(lock_ptr != q->lock_ptr)) {
1009 spin_unlock(lock_ptr);
1012 WARN_ON(plist_node_empty(&q->list));
1013 plist_del(&q->list, &q->list.plist);
1015 BUG_ON(q->pi_state);
1017 spin_unlock(lock_ptr);
1021 drop_futex_key_refs(&q->key);
1026 * PI futexes can not be requeued and must remove themself from the
1027 * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry
1030 static void unqueue_me_pi(struct futex_q *q)
1032 WARN_ON(plist_node_empty(&q->list));
1033 plist_del(&q->list, &q->list.plist);
1035 BUG_ON(!q->pi_state);
1036 free_pi_state(q->pi_state);
1039 spin_unlock(q->lock_ptr);
1041 drop_futex_key_refs(&q->key);
1045 * Fixup the pi_state owner with the new owner.
1047 * Must be called with hash bucket lock held and mm->sem held for non
1050 static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q,
1051 struct task_struct *newowner,
1052 struct rw_semaphore *fshared)
1054 u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS;
1055 struct futex_pi_state *pi_state = q->pi_state;
1056 struct task_struct *oldowner = pi_state->owner;
1057 u32 uval, curval, newval;
1058 int ret, attempt = 0;
1061 if (!pi_state->owner)
1062 newtid |= FUTEX_OWNER_DIED;
1065 * We are here either because we stole the rtmutex from the
1066 * pending owner or we are the pending owner which failed to
1067 * get the rtmutex. We have to replace the pending owner TID
1068 * in the user space variable. This must be atomic as we have
1069 * to preserve the owner died bit here.
1071 * Note: We write the user space value _before_ changing the
1072 * pi_state because we can fault here. Imagine swapped out
1073 * pages or a fork, which was running right before we acquired
1074 * mmap_sem, that marked all the anonymous memory readonly for
1077 * Modifying pi_state _before_ the user space value would
1078 * leave the pi_state in an inconsistent state when we fault
1079 * here, because we need to drop the hash bucket lock to
1080 * handle the fault. This might be observed in the PID check
1081 * in lookup_pi_state.
1084 if (get_futex_value_locked(&uval, uaddr))
1088 newval = (uval & FUTEX_OWNER_DIED) | newtid;
1090 curval = cmpxchg_futex_value_locked(uaddr, uval, newval);
1092 if (curval == -EFAULT)
1100 * We fixed up user space. Now we need to fix the pi_state
1103 if (pi_state->owner != NULL) {
1104 spin_lock_irq(&pi_state->owner->pi_lock);
1105 WARN_ON(list_empty(&pi_state->list));
1106 list_del_init(&pi_state->list);
1107 spin_unlock_irq(&pi_state->owner->pi_lock);
1110 pi_state->owner = newowner;
1112 spin_lock_irq(&newowner->pi_lock);
1113 WARN_ON(!list_empty(&pi_state->list));
1114 list_add(&pi_state->list, &newowner->pi_state_list);
1115 spin_unlock_irq(&newowner->pi_lock);
1119 * To handle the page fault we need to drop the hash bucket
1120 * lock here. That gives the other task (either the pending
1121 * owner itself or the task which stole the rtmutex) the
1122 * chance to try the fixup of the pi_state. So once we are
1123 * back from handling the fault we need to check the pi_state
1124 * after reacquiring the hash bucket lock and before trying to
1125 * do another fixup. When the fixup has been done already we
1129 spin_unlock(q->lock_ptr);
1131 ret = futex_handle_fault((unsigned long)uaddr, fshared, attempt++);
1133 spin_lock(q->lock_ptr);
1136 * Check if someone else fixed it for us:
1138 if (pi_state->owner != oldowner)
1148 * In case we must use restart_block to restart a futex_wait,
1149 * we encode in the 'flags' shared capability
1151 #define FLAGS_SHARED 1
1153 static long futex_wait_restart(struct restart_block *restart);
1155 static int futex_wait(u32 __user *uaddr, struct rw_semaphore *fshared,
1156 u32 val, ktime_t *abs_time, u32 bitset)
1158 struct task_struct *curr = current;
1159 DECLARE_WAITQUEUE(wait, curr);
1160 struct futex_hash_bucket *hb;
1164 struct hrtimer_sleeper t;
1173 q.key = FUTEX_KEY_INIT;
1174 ret = get_futex_key(uaddr, fshared, &q.key);
1175 if (unlikely(ret != 0))
1176 goto out_release_sem;
1178 hb = queue_lock(&q);
1181 * Access the page AFTER the futex is queued.
1182 * Order is important:
1184 * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val);
1185 * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); }
1187 * The basic logical guarantee of a futex is that it blocks ONLY
1188 * if cond(var) is known to be true at the time of blocking, for
1189 * any cond. If we queued after testing *uaddr, that would open
1190 * a race condition where we could block indefinitely with
1191 * cond(var) false, which would violate the guarantee.
1193 * A consequence is that futex_wait() can return zero and absorb
1194 * a wakeup when *uaddr != val on entry to the syscall. This is
1197 * for shared futexes, we hold the mmap semaphore, so the mapping
1198 * cannot have changed since we looked it up in get_futex_key.
1200 ret = get_futex_value_locked(&uval, uaddr);
1202 if (unlikely(ret)) {
1203 queue_unlock(&q, hb);
1205 ret = get_user(uval, uaddr);
1213 goto out_unlock_release_sem;
1215 /* Only actually queue if *uaddr contained val. */
1219 * There might have been scheduling since the queue_me(), as we
1220 * cannot hold a spinlock across the get_user() in case it
1221 * faults, and we cannot just set TASK_INTERRUPTIBLE state when
1222 * queueing ourselves into the futex hash. This code thus has to
1223 * rely on the futex_wake() code removing us from hash when it
1227 /* add_wait_queue is the barrier after __set_current_state. */
1228 __set_current_state(TASK_INTERRUPTIBLE);
1229 add_wait_queue(&q.waiters, &wait);
1231 * !plist_node_empty() is safe here without any lock.
1232 * q.lock_ptr != 0 is not safe, because of ordering against wakeup.
1234 if (likely(!plist_node_empty(&q.list))) {
1238 hrtimer_init_on_stack(&t.timer, CLOCK_MONOTONIC,
1240 hrtimer_init_sleeper(&t, current);
1241 t.timer.expires = *abs_time;
1243 hrtimer_start(&t.timer, t.timer.expires,
1245 if (!hrtimer_active(&t.timer))
1249 * the timer could have already expired, in which
1250 * case current would be flagged for rescheduling.
1251 * Don't bother calling schedule.
1256 hrtimer_cancel(&t.timer);
1258 /* Flag if a timeout occured */
1259 rem = (t.task == NULL);
1261 destroy_hrtimer_on_stack(&t.timer);
1264 __set_current_state(TASK_RUNNING);
1267 * NOTE: we don't remove ourselves from the waitqueue because
1268 * we are the only user of it.
1271 /* If we were woken (and unqueued), we succeeded, whatever. */
1272 if (!unqueue_me(&q))
1278 * We expect signal_pending(current), but another thread may
1279 * have handled it for us already.
1282 return -ERESTARTSYS;
1284 struct restart_block *restart;
1285 restart = ¤t_thread_info()->restart_block;
1286 restart->fn = futex_wait_restart;
1287 restart->futex.uaddr = (u32 *)uaddr;
1288 restart->futex.val = val;
1289 restart->futex.time = abs_time->tv64;
1290 restart->futex.bitset = bitset;
1291 restart->futex.flags = 0;
1294 restart->futex.flags |= FLAGS_SHARED;
1295 return -ERESTART_RESTARTBLOCK;
1298 out_unlock_release_sem:
1299 queue_unlock(&q, hb);
1302 put_futex_key(fshared, &q.key);
1307 static long futex_wait_restart(struct restart_block *restart)
1309 u32 __user *uaddr = (u32 __user *)restart->futex.uaddr;
1310 struct rw_semaphore *fshared = NULL;
1313 t.tv64 = restart->futex.time;
1314 restart->fn = do_no_restart_syscall;
1315 if (restart->futex.flags & FLAGS_SHARED)
1316 fshared = ¤t->mm->mmap_sem;
1317 return (long)futex_wait(uaddr, fshared, restart->futex.val, &t,
1318 restart->futex.bitset);
1323 * Userspace tried a 0 -> TID atomic transition of the futex value
1324 * and failed. The kernel side here does the whole locking operation:
1325 * if there are waiters then it will block, it does PI, etc. (Due to
1326 * races the kernel might see a 0 value of the futex too.)
1328 static int futex_lock_pi(u32 __user *uaddr, struct rw_semaphore *fshared,
1329 int detect, ktime_t *time, int trylock)
1331 struct hrtimer_sleeper timeout, *to = NULL;
1332 struct task_struct *curr = current;
1333 struct futex_hash_bucket *hb;
1334 u32 uval, newval, curval;
1336 int ret, lock_taken, ownerdied = 0, attempt = 0;
1338 if (refill_pi_state_cache())
1343 hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME,
1345 hrtimer_init_sleeper(to, current);
1346 to->timer.expires = *time;
1351 q.key = FUTEX_KEY_INIT;
1352 ret = get_futex_key(uaddr, fshared, &q.key);
1353 if (unlikely(ret != 0))
1354 goto out_release_sem;
1357 hb = queue_lock(&q);
1360 ret = lock_taken = 0;
1363 * To avoid races, we attempt to take the lock here again
1364 * (by doing a 0 -> TID atomic cmpxchg), while holding all
1365 * the locks. It will most likely not succeed.
1367 newval = task_pid_vnr(current);
1369 curval = cmpxchg_futex_value_locked(uaddr, 0, newval);
1371 if (unlikely(curval == -EFAULT))
1375 * Detect deadlocks. In case of REQUEUE_PI this is a valid
1376 * situation and we return success to user space.
1378 if (unlikely((curval & FUTEX_TID_MASK) == task_pid_vnr(current))) {
1380 goto out_unlock_release_sem;
1384 * Surprise - we got the lock. Just return to userspace:
1386 if (unlikely(!curval))
1387 goto out_unlock_release_sem;
1392 * Set the WAITERS flag, so the owner will know it has someone
1393 * to wake at next unlock
1395 newval = curval | FUTEX_WAITERS;
1398 * There are two cases, where a futex might have no owner (the
1399 * owner TID is 0): OWNER_DIED. We take over the futex in this
1400 * case. We also do an unconditional take over, when the owner
1401 * of the futex died.
1403 * This is safe as we are protected by the hash bucket lock !
1405 if (unlikely(ownerdied || !(curval & FUTEX_TID_MASK))) {
1406 /* Keep the OWNER_DIED bit */
1407 newval = (curval & ~FUTEX_TID_MASK) | task_pid_vnr(current);
1412 curval = cmpxchg_futex_value_locked(uaddr, uval, newval);
1414 if (unlikely(curval == -EFAULT))
1416 if (unlikely(curval != uval))
1420 * We took the lock due to owner died take over.
1422 if (unlikely(lock_taken))
1423 goto out_unlock_release_sem;
1426 * We dont have the lock. Look up the PI state (or create it if
1427 * we are the first waiter):
1429 ret = lookup_pi_state(uval, hb, &q.key, &q.pi_state);
1431 if (unlikely(ret)) {
1436 * Task is exiting and we just wait for the
1439 queue_unlock(&q, hb);
1445 * No owner found for this futex. Check if the
1446 * OWNER_DIED bit is set to figure out whether
1447 * this is a robust futex or not.
1449 if (get_futex_value_locked(&curval, uaddr))
1453 * We simply start over in case of a robust
1454 * futex. The code above will take the futex
1457 if (curval & FUTEX_OWNER_DIED) {
1462 goto out_unlock_release_sem;
1467 * Only actually queue now that the atomic ops are done:
1471 WARN_ON(!q.pi_state);
1473 * Block on the PI mutex:
1476 ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1);
1478 ret = rt_mutex_trylock(&q.pi_state->pi_mutex);
1479 /* Fixup the trylock return value: */
1480 ret = ret ? 0 : -EWOULDBLOCK;
1483 spin_lock(q.lock_ptr);
1487 * Got the lock. We might not be the anticipated owner
1488 * if we did a lock-steal - fix up the PI-state in
1491 if (q.pi_state->owner != curr)
1492 ret = fixup_pi_state_owner(uaddr, &q, curr, fshared);
1495 * Catch the rare case, where the lock was released
1496 * when we were on the way back before we locked the
1499 if (q.pi_state->owner == curr) {
1501 * Try to get the rt_mutex now. This might
1502 * fail as some other task acquired the
1503 * rt_mutex after we removed ourself from the
1504 * rt_mutex waiters list.
1506 if (rt_mutex_trylock(&q.pi_state->pi_mutex))
1510 * pi_state is incorrect, some other
1511 * task did a lock steal and we
1512 * returned due to timeout or signal
1513 * without taking the rt_mutex. Too
1514 * late. We can access the
1515 * rt_mutex_owner without locking, as
1516 * the other task is now blocked on
1517 * the hash bucket lock. Fix the state
1520 struct task_struct *owner;
1523 owner = rt_mutex_owner(&q.pi_state->pi_mutex);
1524 res = fixup_pi_state_owner(uaddr, &q, owner,
1527 /* propagate -EFAULT, if the fixup failed */
1533 * Paranoia check. If we did not take the lock
1534 * in the trylock above, then we should not be
1535 * the owner of the rtmutex, neither the real
1536 * nor the pending one:
1538 if (rt_mutex_owner(&q.pi_state->pi_mutex) == curr)
1539 printk(KERN_ERR "futex_lock_pi: ret = %d "
1540 "pi-mutex: %p pi-state %p\n", ret,
1541 q.pi_state->pi_mutex.owner,
1546 /* Unqueue and drop the lock */
1550 destroy_hrtimer_on_stack(&to->timer);
1551 return ret != -EINTR ? ret : -ERESTARTNOINTR;
1553 out_unlock_release_sem:
1554 queue_unlock(&q, hb);
1557 put_futex_key(fshared, &q.key);
1559 destroy_hrtimer_on_stack(&to->timer);
1564 * We have to r/w *(int __user *)uaddr, but we can't modify it
1565 * non-atomically. Therefore, if get_user below is not
1566 * enough, we need to handle the fault ourselves, while
1567 * still holding the mmap_sem.
1569 * ... and hb->lock. :-) --ANK
1571 queue_unlock(&q, hb);
1574 ret = futex_handle_fault((unsigned long)uaddr, fshared,
1577 goto out_release_sem;
1578 goto retry_unlocked;
1581 ret = get_user(uval, uaddr);
1582 if (!ret && (uval != -EFAULT))
1586 destroy_hrtimer_on_stack(&to->timer);
1591 * Userspace attempted a TID -> 0 atomic transition, and failed.
1592 * This is the in-kernel slowpath: we look up the PI state (if any),
1593 * and do the rt-mutex unlock.
1595 static int futex_unlock_pi(u32 __user *uaddr, struct rw_semaphore *fshared)
1597 struct futex_hash_bucket *hb;
1598 struct futex_q *this, *next;
1600 struct plist_head *head;
1601 union futex_key key = FUTEX_KEY_INIT;
1602 int ret, attempt = 0;
1605 if (get_user(uval, uaddr))
1608 * We release only a lock we actually own:
1610 if ((uval & FUTEX_TID_MASK) != task_pid_vnr(current))
1613 ret = get_futex_key(uaddr, fshared, &key);
1614 if (unlikely(ret != 0))
1617 hb = hash_futex(&key);
1619 spin_lock(&hb->lock);
1622 * To avoid races, try to do the TID -> 0 atomic transition
1623 * again. If it succeeds then we can return without waking
1626 if (!(uval & FUTEX_OWNER_DIED))
1627 uval = cmpxchg_futex_value_locked(uaddr, task_pid_vnr(current), 0);
1630 if (unlikely(uval == -EFAULT))
1633 * Rare case: we managed to release the lock atomically,
1634 * no need to wake anyone else up:
1636 if (unlikely(uval == task_pid_vnr(current)))
1640 * Ok, other tasks may need to be woken up - check waiters
1641 * and do the wakeup if necessary:
1645 plist_for_each_entry_safe(this, next, head, list) {
1646 if (!match_futex (&this->key, &key))
1648 ret = wake_futex_pi(uaddr, uval, this);
1650 * The atomic access to the futex value
1651 * generated a pagefault, so retry the
1652 * user-access and the wakeup:
1659 * No waiters - kernel unlocks the futex:
1661 if (!(uval & FUTEX_OWNER_DIED)) {
1662 ret = unlock_futex_pi(uaddr, uval);
1668 spin_unlock(&hb->lock);
1670 put_futex_key(fshared, &key);
1676 * We have to r/w *(int __user *)uaddr, but we can't modify it
1677 * non-atomically. Therefore, if get_user below is not
1678 * enough, we need to handle the fault ourselves, while
1679 * still holding the mmap_sem.
1681 * ... and hb->lock. --ANK
1683 spin_unlock(&hb->lock);
1686 ret = futex_handle_fault((unsigned long)uaddr, fshared,
1691 goto retry_unlocked;
1694 ret = get_user(uval, uaddr);
1695 if (!ret && (uval != -EFAULT))
1702 * Support for robust futexes: the kernel cleans up held futexes at
1705 * Implementation: user-space maintains a per-thread list of locks it
1706 * is holding. Upon do_exit(), the kernel carefully walks this list,
1707 * and marks all locks that are owned by this thread with the
1708 * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is
1709 * always manipulated with the lock held, so the list is private and
1710 * per-thread. Userspace also maintains a per-thread 'list_op_pending'
1711 * field, to allow the kernel to clean up if the thread dies after
1712 * acquiring the lock, but just before it could have added itself to
1713 * the list. There can only be one such pending lock.
1717 * sys_set_robust_list - set the robust-futex list head of a task
1718 * @head: pointer to the list-head
1719 * @len: length of the list-head, as userspace expects
1722 sys_set_robust_list(struct robust_list_head __user *head,
1725 if (!futex_cmpxchg_enabled)
1728 * The kernel knows only one size for now:
1730 if (unlikely(len != sizeof(*head)))
1733 current->robust_list = head;
1739 * sys_get_robust_list - get the robust-futex list head of a task
1740 * @pid: pid of the process [zero for current task]
1741 * @head_ptr: pointer to a list-head pointer, the kernel fills it in
1742 * @len_ptr: pointer to a length field, the kernel fills in the header size
1745 sys_get_robust_list(int pid, struct robust_list_head __user * __user *head_ptr,
1746 size_t __user *len_ptr)
1748 struct robust_list_head __user *head;
1751 if (!futex_cmpxchg_enabled)
1755 head = current->robust_list;
1757 struct task_struct *p;
1761 p = find_task_by_vpid(pid);
1765 if ((current->euid != p->euid) && (current->euid != p->uid) &&
1766 !capable(CAP_SYS_PTRACE))
1768 head = p->robust_list;
1772 if (put_user(sizeof(*head), len_ptr))
1774 return put_user(head, head_ptr);
1783 * Process a futex-list entry, check whether it's owned by the
1784 * dying task, and do notification if so:
1786 int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi)
1788 u32 uval, nval, mval;
1791 if (get_user(uval, uaddr))
1794 if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) {
1796 * Ok, this dying thread is truly holding a futex
1797 * of interest. Set the OWNER_DIED bit atomically
1798 * via cmpxchg, and if the value had FUTEX_WAITERS
1799 * set, wake up a waiter (if any). (We have to do a
1800 * futex_wake() even if OWNER_DIED is already set -
1801 * to handle the rare but possible case of recursive
1802 * thread-death.) The rest of the cleanup is done in
1805 mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED;
1806 nval = futex_atomic_cmpxchg_inatomic(uaddr, uval, mval);
1808 if (nval == -EFAULT)
1815 * Wake robust non-PI futexes here. The wakeup of
1816 * PI futexes happens in exit_pi_state():
1818 if (!pi && (uval & FUTEX_WAITERS))
1819 futex_wake(uaddr, &curr->mm->mmap_sem, 1,
1820 FUTEX_BITSET_MATCH_ANY);
1826 * Fetch a robust-list pointer. Bit 0 signals PI futexes:
1828 static inline int fetch_robust_entry(struct robust_list __user **entry,
1829 struct robust_list __user * __user *head,
1832 unsigned long uentry;
1834 if (get_user(uentry, (unsigned long __user *)head))
1837 *entry = (void __user *)(uentry & ~1UL);
1844 * Walk curr->robust_list (very carefully, it's a userspace list!)
1845 * and mark any locks found there dead, and notify any waiters.
1847 * We silently return on any sign of list-walking problem.
1849 void exit_robust_list(struct task_struct *curr)
1851 struct robust_list_head __user *head = curr->robust_list;
1852 struct robust_list __user *entry, *next_entry, *pending;
1853 unsigned int limit = ROBUST_LIST_LIMIT, pi, next_pi, pip;
1854 unsigned long futex_offset;
1857 if (!futex_cmpxchg_enabled)
1861 * Fetch the list head (which was registered earlier, via
1862 * sys_set_robust_list()):
1864 if (fetch_robust_entry(&entry, &head->list.next, &pi))
1867 * Fetch the relative futex offset:
1869 if (get_user(futex_offset, &head->futex_offset))
1872 * Fetch any possibly pending lock-add first, and handle it
1875 if (fetch_robust_entry(&pending, &head->list_op_pending, &pip))
1878 next_entry = NULL; /* avoid warning with gcc */
1879 while (entry != &head->list) {
1881 * Fetch the next entry in the list before calling
1882 * handle_futex_death:
1884 rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi);
1886 * A pending lock might already be on the list, so
1887 * don't process it twice:
1889 if (entry != pending)
1890 if (handle_futex_death((void __user *)entry + futex_offset,
1898 * Avoid excessively long or circular lists:
1907 handle_futex_death((void __user *)pending + futex_offset,
1911 long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout,
1912 u32 __user *uaddr2, u32 val2, u32 val3)
1915 int cmd = op & FUTEX_CMD_MASK;
1916 struct rw_semaphore *fshared = NULL;
1918 if (!(op & FUTEX_PRIVATE_FLAG))
1919 fshared = ¤t->mm->mmap_sem;
1923 val3 = FUTEX_BITSET_MATCH_ANY;
1924 case FUTEX_WAIT_BITSET:
1925 ret = futex_wait(uaddr, fshared, val, timeout, val3);
1928 val3 = FUTEX_BITSET_MATCH_ANY;
1929 case FUTEX_WAKE_BITSET:
1930 ret = futex_wake(uaddr, fshared, val, val3);
1933 ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, NULL);
1935 case FUTEX_CMP_REQUEUE:
1936 ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, &val3);
1939 ret = futex_wake_op(uaddr, fshared, uaddr2, val, val2, val3);
1942 if (futex_cmpxchg_enabled)
1943 ret = futex_lock_pi(uaddr, fshared, val, timeout, 0);
1945 case FUTEX_UNLOCK_PI:
1946 if (futex_cmpxchg_enabled)
1947 ret = futex_unlock_pi(uaddr, fshared);
1949 case FUTEX_TRYLOCK_PI:
1950 if (futex_cmpxchg_enabled)
1951 ret = futex_lock_pi(uaddr, fshared, 0, timeout, 1);
1960 asmlinkage long sys_futex(u32 __user *uaddr, int op, u32 val,
1961 struct timespec __user *utime, u32 __user *uaddr2,
1965 ktime_t t, *tp = NULL;
1967 int cmd = op & FUTEX_CMD_MASK;
1969 if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI ||
1970 cmd == FUTEX_WAIT_BITSET)) {
1971 if (copy_from_user(&ts, utime, sizeof(ts)) != 0)
1973 if (!timespec_valid(&ts))
1976 t = timespec_to_ktime(ts);
1977 if (cmd == FUTEX_WAIT)
1978 t = ktime_add_safe(ktime_get(), t);
1982 * requeue parameter in 'utime' if cmd == FUTEX_REQUEUE.
1983 * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP.
1985 if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE ||
1986 cmd == FUTEX_WAKE_OP)
1987 val2 = (u32) (unsigned long) utime;
1989 return do_futex(uaddr, op, val, tp, uaddr2, val2, val3);
1992 static int __init futex_init(void)
1998 * This will fail and we want it. Some arch implementations do
1999 * runtime detection of the futex_atomic_cmpxchg_inatomic()
2000 * functionality. We want to know that before we call in any
2001 * of the complex code paths. Also we want to prevent
2002 * registration of robust lists in that case. NULL is
2003 * guaranteed to fault and we get -EFAULT on functional
2004 * implementation, the non functional ones will return
2007 curval = cmpxchg_futex_value_locked(NULL, 0, 0);
2008 if (curval == -EFAULT)
2009 futex_cmpxchg_enabled = 1;
2011 for (i = 0; i < ARRAY_SIZE(futex_queues); i++) {
2012 plist_head_init(&futex_queues[i].chain, &futex_queues[i].lock);
2013 spin_lock_init(&futex_queues[i].lock);
2018 __initcall(futex_init);