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 * Hash buckets are shared by all the futex_keys that hash to the same
118 * location. Each key may have multiple futex_q structures, one for each task
119 * waiting on a futex.
121 struct futex_hash_bucket {
123 struct plist_head chain;
126 static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS];
129 * We hash on the keys returned from get_futex_key (see below).
131 static struct futex_hash_bucket *hash_futex(union futex_key *key)
133 u32 hash = jhash2((u32*)&key->both.word,
134 (sizeof(key->both.word)+sizeof(key->both.ptr))/4,
136 return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)];
140 * Return 1 if two futex_keys are equal, 0 otherwise.
142 static inline int match_futex(union futex_key *key1, union futex_key *key2)
144 return (key1->both.word == key2->both.word
145 && key1->both.ptr == key2->both.ptr
146 && key1->both.offset == key2->both.offset);
150 * Take a reference to the resource addressed by a key.
151 * Can be called while holding spinlocks.
154 static void get_futex_key_refs(union futex_key *key)
159 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
161 atomic_inc(&key->shared.inode->i_count);
163 case FUT_OFF_MMSHARED:
164 atomic_inc(&key->private.mm->mm_count);
170 * Drop a reference to the resource addressed by a key.
171 * The hash bucket spinlock must not be held.
173 static void drop_futex_key_refs(union futex_key *key)
175 if (!key->both.ptr) {
176 /* If we're here then we tried to put a key we failed to get */
181 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
183 iput(key->shared.inode);
185 case FUT_OFF_MMSHARED:
186 mmdrop(key->private.mm);
192 * get_futex_key - Get parameters which are the keys for a futex.
193 * @uaddr: virtual address of the futex
194 * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED
195 * @key: address where result is stored.
197 * Returns a negative error code or 0
198 * The key words are stored in *key on success.
200 * For shared mappings, it's (page->index, vma->vm_file->f_path.dentry->d_inode,
201 * offset_within_page). For private mappings, it's (uaddr, current->mm).
202 * We can usually work out the index without swapping in the page.
204 * lock_page() might sleep, the caller should not hold a spinlock.
206 static int get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key)
208 unsigned long address = (unsigned long)uaddr;
209 struct mm_struct *mm = current->mm;
214 * The futex address must be "naturally" aligned.
216 key->both.offset = address % PAGE_SIZE;
217 if (unlikely((address % sizeof(u32)) != 0))
219 address -= key->both.offset;
222 * PROCESS_PRIVATE futexes are fast.
223 * As the mm cannot disappear under us and the 'key' only needs
224 * virtual address, we dont even have to find the underlying vma.
225 * Note : We do have to check 'uaddr' is a valid user address,
226 * but access_ok() should be faster than find_vma()
229 if (unlikely(!access_ok(VERIFY_WRITE, uaddr, sizeof(u32))))
231 key->private.mm = mm;
232 key->private.address = address;
233 get_futex_key_refs(key);
238 err = get_user_pages_fast(address, 1, 0, &page);
243 if (!page->mapping) {
250 * Private mappings are handled in a simple way.
252 * NOTE: When userspace waits on a MAP_SHARED mapping, even if
253 * it's a read-only handle, it's expected that futexes attach to
254 * the object not the particular process.
256 if (PageAnon(page)) {
257 key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */
258 key->private.mm = mm;
259 key->private.address = address;
261 key->both.offset |= FUT_OFF_INODE; /* inode-based key */
262 key->shared.inode = page->mapping->host;
263 key->shared.pgoff = page->index;
266 get_futex_key_refs(key);
274 void put_futex_key(int fshared, union futex_key *key)
276 drop_futex_key_refs(key);
279 static u32 cmpxchg_futex_value_locked(u32 __user *uaddr, u32 uval, u32 newval)
284 curval = futex_atomic_cmpxchg_inatomic(uaddr, uval, newval);
290 static int get_futex_value_locked(u32 *dest, u32 __user *from)
295 ret = __copy_from_user_inatomic(dest, from, sizeof(u32));
298 return ret ? -EFAULT : 0;
304 static int futex_handle_fault(unsigned long address, int attempt)
306 struct vm_area_struct * vma;
307 struct mm_struct *mm = current->mm;
313 down_read(&mm->mmap_sem);
314 vma = find_vma(mm, address);
315 if (vma && address >= vma->vm_start &&
316 (vma->vm_flags & VM_WRITE)) {
318 fault = handle_mm_fault(mm, vma, address, 1);
319 if (unlikely((fault & VM_FAULT_ERROR))) {
321 /* XXX: let's do this when we verify it is OK */
322 if (ret & VM_FAULT_OOM)
327 if (fault & VM_FAULT_MAJOR)
333 up_read(&mm->mmap_sem);
340 static int refill_pi_state_cache(void)
342 struct futex_pi_state *pi_state;
344 if (likely(current->pi_state_cache))
347 pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL);
352 INIT_LIST_HEAD(&pi_state->list);
353 /* pi_mutex gets initialized later */
354 pi_state->owner = NULL;
355 atomic_set(&pi_state->refcount, 1);
356 pi_state->key = FUTEX_KEY_INIT;
358 current->pi_state_cache = pi_state;
363 static struct futex_pi_state * alloc_pi_state(void)
365 struct futex_pi_state *pi_state = current->pi_state_cache;
368 current->pi_state_cache = NULL;
373 static void free_pi_state(struct futex_pi_state *pi_state)
375 if (!atomic_dec_and_test(&pi_state->refcount))
379 * If pi_state->owner is NULL, the owner is most probably dying
380 * and has cleaned up the pi_state already
382 if (pi_state->owner) {
383 spin_lock_irq(&pi_state->owner->pi_lock);
384 list_del_init(&pi_state->list);
385 spin_unlock_irq(&pi_state->owner->pi_lock);
387 rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner);
390 if (current->pi_state_cache)
394 * pi_state->list is already empty.
395 * clear pi_state->owner.
396 * refcount is at 0 - put it back to 1.
398 pi_state->owner = NULL;
399 atomic_set(&pi_state->refcount, 1);
400 current->pi_state_cache = pi_state;
405 * Look up the task based on what TID userspace gave us.
408 static struct task_struct * futex_find_get_task(pid_t pid)
410 struct task_struct *p;
411 const struct cred *cred = current_cred(), *pcred;
414 p = find_task_by_vpid(pid);
418 pcred = __task_cred(p);
419 if (cred->euid != pcred->euid &&
420 cred->euid != pcred->uid)
432 * This task is holding PI mutexes at exit time => bad.
433 * Kernel cleans up PI-state, but userspace is likely hosed.
434 * (Robust-futex cleanup is separate and might save the day for userspace.)
436 void exit_pi_state_list(struct task_struct *curr)
438 struct list_head *next, *head = &curr->pi_state_list;
439 struct futex_pi_state *pi_state;
440 struct futex_hash_bucket *hb;
441 union futex_key key = FUTEX_KEY_INIT;
443 if (!futex_cmpxchg_enabled)
446 * We are a ZOMBIE and nobody can enqueue itself on
447 * pi_state_list anymore, but we have to be careful
448 * versus waiters unqueueing themselves:
450 spin_lock_irq(&curr->pi_lock);
451 while (!list_empty(head)) {
454 pi_state = list_entry(next, struct futex_pi_state, list);
456 hb = hash_futex(&key);
457 spin_unlock_irq(&curr->pi_lock);
459 spin_lock(&hb->lock);
461 spin_lock_irq(&curr->pi_lock);
463 * We dropped the pi-lock, so re-check whether this
464 * task still owns the PI-state:
466 if (head->next != next) {
467 spin_unlock(&hb->lock);
471 WARN_ON(pi_state->owner != curr);
472 WARN_ON(list_empty(&pi_state->list));
473 list_del_init(&pi_state->list);
474 pi_state->owner = NULL;
475 spin_unlock_irq(&curr->pi_lock);
477 rt_mutex_unlock(&pi_state->pi_mutex);
479 spin_unlock(&hb->lock);
481 spin_lock_irq(&curr->pi_lock);
483 spin_unlock_irq(&curr->pi_lock);
487 lookup_pi_state(u32 uval, struct futex_hash_bucket *hb,
488 union futex_key *key, struct futex_pi_state **ps)
490 struct futex_pi_state *pi_state = NULL;
491 struct futex_q *this, *next;
492 struct plist_head *head;
493 struct task_struct *p;
494 pid_t pid = uval & FUTEX_TID_MASK;
498 plist_for_each_entry_safe(this, next, head, list) {
499 if (match_futex(&this->key, key)) {
501 * Another waiter already exists - bump up
502 * the refcount and return its pi_state:
504 pi_state = this->pi_state;
506 * Userspace might have messed up non PI and PI futexes
508 if (unlikely(!pi_state))
511 WARN_ON(!atomic_read(&pi_state->refcount));
512 WARN_ON(pid && pi_state->owner &&
513 pi_state->owner->pid != pid);
515 atomic_inc(&pi_state->refcount);
523 * We are the first waiter - try to look up the real owner and attach
524 * the new pi_state to it, but bail out when TID = 0
528 p = futex_find_get_task(pid);
533 * We need to look at the task state flags to figure out,
534 * whether the task is exiting. To protect against the do_exit
535 * change of the task flags, we do this protected by
538 spin_lock_irq(&p->pi_lock);
539 if (unlikely(p->flags & PF_EXITING)) {
541 * The task is on the way out. When PF_EXITPIDONE is
542 * set, we know that the task has finished the
545 int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN;
547 spin_unlock_irq(&p->pi_lock);
552 pi_state = alloc_pi_state();
555 * Initialize the pi_mutex in locked state and make 'p'
558 rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p);
560 /* Store the key for possible exit cleanups: */
561 pi_state->key = *key;
563 WARN_ON(!list_empty(&pi_state->list));
564 list_add(&pi_state->list, &p->pi_state_list);
566 spin_unlock_irq(&p->pi_lock);
576 * The hash bucket lock must be held when this is called.
577 * Afterwards, the futex_q must not be accessed.
579 static void wake_futex(struct futex_q *q)
581 plist_del(&q->list, &q->list.plist);
583 * The lock in wake_up_all() is a crucial memory barrier after the
584 * plist_del() and also before assigning to q->lock_ptr.
588 * The waiting task can free the futex_q as soon as this is written,
589 * without taking any locks. This must come last.
591 * A memory barrier is required here to prevent the following store to
592 * lock_ptr from getting ahead of the wakeup. Clearing the lock at the
593 * end of wake_up() does not prevent this store from moving.
599 static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this)
601 struct task_struct *new_owner;
602 struct futex_pi_state *pi_state = this->pi_state;
608 spin_lock(&pi_state->pi_mutex.wait_lock);
609 new_owner = rt_mutex_next_owner(&pi_state->pi_mutex);
612 * This happens when we have stolen the lock and the original
613 * pending owner did not enqueue itself back on the rt_mutex.
614 * Thats not a tragedy. We know that way, that a lock waiter
615 * is on the fly. We make the futex_q waiter the pending owner.
618 new_owner = this->task;
621 * We pass it to the next owner. (The WAITERS bit is always
622 * kept enabled while there is PI state around. We must also
623 * preserve the owner died bit.)
625 if (!(uval & FUTEX_OWNER_DIED)) {
628 newval = FUTEX_WAITERS | task_pid_vnr(new_owner);
630 curval = cmpxchg_futex_value_locked(uaddr, uval, newval);
632 if (curval == -EFAULT)
634 else if (curval != uval)
637 spin_unlock(&pi_state->pi_mutex.wait_lock);
642 spin_lock_irq(&pi_state->owner->pi_lock);
643 WARN_ON(list_empty(&pi_state->list));
644 list_del_init(&pi_state->list);
645 spin_unlock_irq(&pi_state->owner->pi_lock);
647 spin_lock_irq(&new_owner->pi_lock);
648 WARN_ON(!list_empty(&pi_state->list));
649 list_add(&pi_state->list, &new_owner->pi_state_list);
650 pi_state->owner = new_owner;
651 spin_unlock_irq(&new_owner->pi_lock);
653 spin_unlock(&pi_state->pi_mutex.wait_lock);
654 rt_mutex_unlock(&pi_state->pi_mutex);
659 static int unlock_futex_pi(u32 __user *uaddr, u32 uval)
664 * There is no waiter, so we unlock the futex. The owner died
665 * bit has not to be preserved here. We are the owner:
667 oldval = cmpxchg_futex_value_locked(uaddr, uval, 0);
669 if (oldval == -EFAULT)
678 * Express the locking dependencies for lockdep:
681 double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
684 spin_lock(&hb1->lock);
686 spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING);
687 } else { /* hb1 > hb2 */
688 spin_lock(&hb2->lock);
689 spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING);
694 double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
697 spin_unlock(&hb2->lock);
699 spin_unlock(&hb1->lock);
700 } else { /* hb1 > hb2 */
701 spin_unlock(&hb1->lock);
702 spin_unlock(&hb2->lock);
707 * Wake up waiters matching bitset queued on this futex (uaddr).
709 static int futex_wake(u32 __user *uaddr, int fshared, int nr_wake, u32 bitset)
711 struct futex_hash_bucket *hb;
712 struct futex_q *this, *next;
713 struct plist_head *head;
714 union futex_key key = FUTEX_KEY_INIT;
720 ret = get_futex_key(uaddr, fshared, &key);
721 if (unlikely(ret != 0))
724 hb = hash_futex(&key);
725 spin_lock(&hb->lock);
728 plist_for_each_entry_safe(this, next, head, list) {
729 if (match_futex (&this->key, &key)) {
730 if (this->pi_state) {
735 /* Check if one of the bits is set in both bitsets */
736 if (!(this->bitset & bitset))
740 if (++ret >= nr_wake)
745 spin_unlock(&hb->lock);
746 put_futex_key(fshared, &key);
752 * Wake up all waiters hashed on the physical page that is mapped
753 * to this virtual address:
756 futex_wake_op(u32 __user *uaddr1, int fshared, u32 __user *uaddr2,
757 int nr_wake, int nr_wake2, int op)
759 union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
760 struct futex_hash_bucket *hb1, *hb2;
761 struct plist_head *head;
762 struct futex_q *this, *next;
763 int ret, op_ret, attempt = 0;
766 ret = get_futex_key(uaddr1, fshared, &key1);
767 if (unlikely(ret != 0))
769 ret = get_futex_key(uaddr2, fshared, &key2);
770 if (unlikely(ret != 0))
773 hb1 = hash_futex(&key1);
774 hb2 = hash_futex(&key2);
777 double_lock_hb(hb1, hb2);
779 op_ret = futex_atomic_op_inuser(op, uaddr2);
780 if (unlikely(op_ret < 0)) {
783 double_unlock_hb(hb1, hb2);
787 * we don't get EFAULT from MMU faults if we don't have an MMU,
788 * but we might get them from range checking
794 if (unlikely(op_ret != -EFAULT)) {
800 * futex_atomic_op_inuser needs to both read and write
801 * *(int __user *)uaddr2, but we can't modify it
802 * non-atomically. Therefore, if get_user below is not
803 * enough, we need to handle the fault ourselves, while
804 * still holding the mmap_sem.
807 ret = futex_handle_fault((unsigned long)uaddr2,
814 ret = get_user(dummy, uaddr2);
818 put_futex_key(fshared, &key2);
819 put_futex_key(fshared, &key1);
825 plist_for_each_entry_safe(this, next, head, list) {
826 if (match_futex (&this->key, &key1)) {
828 if (++ret >= nr_wake)
837 plist_for_each_entry_safe(this, next, head, list) {
838 if (match_futex (&this->key, &key2)) {
840 if (++op_ret >= nr_wake2)
847 double_unlock_hb(hb1, hb2);
849 put_futex_key(fshared, &key2);
851 put_futex_key(fshared, &key1);
857 * Requeue all waiters hashed on one physical page to another
860 static int futex_requeue(u32 __user *uaddr1, int fshared, u32 __user *uaddr2,
861 int nr_wake, int nr_requeue, u32 *cmpval)
863 union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
864 struct futex_hash_bucket *hb1, *hb2;
865 struct plist_head *head1;
866 struct futex_q *this, *next;
867 int ret, drop_count = 0;
870 ret = get_futex_key(uaddr1, fshared, &key1);
871 if (unlikely(ret != 0))
873 ret = get_futex_key(uaddr2, fshared, &key2);
874 if (unlikely(ret != 0))
877 hb1 = hash_futex(&key1);
878 hb2 = hash_futex(&key2);
880 double_lock_hb(hb1, hb2);
882 if (likely(cmpval != NULL)) {
885 ret = get_futex_value_locked(&curval, uaddr1);
888 double_unlock_hb(hb1, hb2);
890 put_futex_key(fshared, &key2);
891 put_futex_key(fshared, &key1);
893 ret = get_user(curval, uaddr1);
900 if (curval != *cmpval) {
907 plist_for_each_entry_safe(this, next, head1, list) {
908 if (!match_futex (&this->key, &key1))
910 if (++ret <= nr_wake) {
914 * If key1 and key2 hash to the same bucket, no need to
917 if (likely(head1 != &hb2->chain)) {
918 plist_del(&this->list, &hb1->chain);
919 plist_add(&this->list, &hb2->chain);
920 this->lock_ptr = &hb2->lock;
921 #ifdef CONFIG_DEBUG_PI_LIST
922 this->list.plist.lock = &hb2->lock;
926 get_futex_key_refs(&key2);
929 if (ret - nr_wake >= nr_requeue)
935 double_unlock_hb(hb1, hb2);
937 /* drop_futex_key_refs() must be called outside the spinlocks. */
938 while (--drop_count >= 0)
939 drop_futex_key_refs(&key1);
942 put_futex_key(fshared, &key2);
944 put_futex_key(fshared, &key1);
949 /* The key must be already stored in q->key. */
950 static inline struct futex_hash_bucket *queue_lock(struct futex_q *q)
952 struct futex_hash_bucket *hb;
954 init_waitqueue_head(&q->waiter);
956 get_futex_key_refs(&q->key);
957 hb = hash_futex(&q->key);
958 q->lock_ptr = &hb->lock;
960 spin_lock(&hb->lock);
964 static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb)
969 * The priority used to register this element is
970 * - either the real thread-priority for the real-time threads
971 * (i.e. threads with a priority lower than MAX_RT_PRIO)
972 * - or MAX_RT_PRIO for non-RT threads.
973 * Thus, all RT-threads are woken first in priority order, and
974 * the others are woken last, in FIFO order.
976 prio = min(current->normal_prio, MAX_RT_PRIO);
978 plist_node_init(&q->list, prio);
979 #ifdef CONFIG_DEBUG_PI_LIST
980 q->list.plist.lock = &hb->lock;
982 plist_add(&q->list, &hb->chain);
984 spin_unlock(&hb->lock);
988 queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb)
990 spin_unlock(&hb->lock);
991 drop_futex_key_refs(&q->key);
995 * queue_me and unqueue_me must be called as a pair, each
996 * exactly once. They are called with the hashed spinlock held.
999 /* Return 1 if we were still queued (ie. 0 means we were woken) */
1000 static int unqueue_me(struct futex_q *q)
1002 spinlock_t *lock_ptr;
1005 /* In the common case we don't take the spinlock, which is nice. */
1007 lock_ptr = q->lock_ptr;
1009 if (lock_ptr != NULL) {
1010 spin_lock(lock_ptr);
1012 * q->lock_ptr can change between reading it and
1013 * spin_lock(), causing us to take the wrong lock. This
1014 * corrects the race condition.
1016 * Reasoning goes like this: if we have the wrong lock,
1017 * q->lock_ptr must have changed (maybe several times)
1018 * between reading it and the spin_lock(). It can
1019 * change again after the spin_lock() but only if it was
1020 * already changed before the spin_lock(). It cannot,
1021 * however, change back to the original value. Therefore
1022 * we can detect whether we acquired the correct lock.
1024 if (unlikely(lock_ptr != q->lock_ptr)) {
1025 spin_unlock(lock_ptr);
1028 WARN_ON(plist_node_empty(&q->list));
1029 plist_del(&q->list, &q->list.plist);
1031 BUG_ON(q->pi_state);
1033 spin_unlock(lock_ptr);
1037 drop_futex_key_refs(&q->key);
1042 * PI futexes can not be requeued and must remove themself from the
1043 * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry
1046 static void unqueue_me_pi(struct futex_q *q)
1048 WARN_ON(plist_node_empty(&q->list));
1049 plist_del(&q->list, &q->list.plist);
1051 BUG_ON(!q->pi_state);
1052 free_pi_state(q->pi_state);
1055 spin_unlock(q->lock_ptr);
1057 drop_futex_key_refs(&q->key);
1061 * Fixup the pi_state owner with the new owner.
1063 * Must be called with hash bucket lock held and mm->sem held for non
1066 static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q,
1067 struct task_struct *newowner, int fshared)
1069 u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS;
1070 struct futex_pi_state *pi_state = q->pi_state;
1071 struct task_struct *oldowner = pi_state->owner;
1072 u32 uval, curval, newval;
1073 int ret, attempt = 0;
1076 if (!pi_state->owner)
1077 newtid |= FUTEX_OWNER_DIED;
1080 * We are here either because we stole the rtmutex from the
1081 * pending owner or we are the pending owner which failed to
1082 * get the rtmutex. We have to replace the pending owner TID
1083 * in the user space variable. This must be atomic as we have
1084 * to preserve the owner died bit here.
1086 * Note: We write the user space value _before_ changing the pi_state
1087 * because we can fault here. Imagine swapped out pages or a fork
1088 * that marked all the anonymous memory readonly for cow.
1090 * Modifying pi_state _before_ the user space value would
1091 * leave the pi_state in an inconsistent state when we fault
1092 * here, because we need to drop the hash bucket lock to
1093 * handle the fault. This might be observed in the PID check
1094 * in lookup_pi_state.
1097 if (get_futex_value_locked(&uval, uaddr))
1101 newval = (uval & FUTEX_OWNER_DIED) | newtid;
1103 curval = cmpxchg_futex_value_locked(uaddr, uval, newval);
1105 if (curval == -EFAULT)
1113 * We fixed up user space. Now we need to fix the pi_state
1116 if (pi_state->owner != NULL) {
1117 spin_lock_irq(&pi_state->owner->pi_lock);
1118 WARN_ON(list_empty(&pi_state->list));
1119 list_del_init(&pi_state->list);
1120 spin_unlock_irq(&pi_state->owner->pi_lock);
1123 pi_state->owner = newowner;
1125 spin_lock_irq(&newowner->pi_lock);
1126 WARN_ON(!list_empty(&pi_state->list));
1127 list_add(&pi_state->list, &newowner->pi_state_list);
1128 spin_unlock_irq(&newowner->pi_lock);
1132 * To handle the page fault we need to drop the hash bucket
1133 * lock here. That gives the other task (either the pending
1134 * owner itself or the task which stole the rtmutex) the
1135 * chance to try the fixup of the pi_state. So once we are
1136 * back from handling the fault we need to check the pi_state
1137 * after reacquiring the hash bucket lock and before trying to
1138 * do another fixup. When the fixup has been done already we
1142 spin_unlock(q->lock_ptr);
1144 ret = futex_handle_fault((unsigned long)uaddr, attempt++);
1146 spin_lock(q->lock_ptr);
1149 * Check if someone else fixed it for us:
1151 if (pi_state->owner != oldowner)
1161 * In case we must use restart_block to restart a futex_wait,
1162 * we encode in the 'flags' shared capability
1164 #define FLAGS_SHARED 0x01
1165 #define FLAGS_CLOCKRT 0x02
1167 static long futex_wait_restart(struct restart_block *restart);
1169 static int futex_wait(u32 __user *uaddr, int fshared,
1170 u32 val, ktime_t *abs_time, u32 bitset, int clockrt)
1172 struct task_struct *curr = current;
1173 struct restart_block *restart;
1174 DECLARE_WAITQUEUE(wait, curr);
1175 struct futex_hash_bucket *hb;
1179 struct hrtimer_sleeper t;
1188 q.key = FUTEX_KEY_INIT;
1189 ret = get_futex_key(uaddr, fshared, &q.key);
1190 if (unlikely(ret != 0))
1193 hb = queue_lock(&q);
1196 * Access the page AFTER the hash-bucket is locked.
1197 * Order is important:
1199 * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val);
1200 * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); }
1202 * The basic logical guarantee of a futex is that it blocks ONLY
1203 * if cond(var) is known to be true at the time of blocking, for
1204 * any cond. If we queued after testing *uaddr, that would open
1205 * a race condition where we could block indefinitely with
1206 * cond(var) false, which would violate the guarantee.
1208 * A consequence is that futex_wait() can return zero and absorb
1209 * a wakeup when *uaddr != val on entry to the syscall. This is
1212 * For shared futexes, we hold the mmap semaphore, so the mapping
1213 * cannot have changed since we looked it up in get_futex_key.
1215 ret = get_futex_value_locked(&uval, uaddr);
1217 if (unlikely(ret)) {
1218 queue_unlock(&q, hb);
1219 put_futex_key(fshared, &q.key);
1221 ret = get_user(uval, uaddr);
1228 if (unlikely(uval != val)) {
1229 queue_unlock(&q, hb);
1233 /* Only actually queue if *uaddr contained val. */
1237 * There might have been scheduling since the queue_me(), as we
1238 * cannot hold a spinlock across the get_user() in case it
1239 * faults, and we cannot just set TASK_INTERRUPTIBLE state when
1240 * queueing ourselves into the futex hash. This code thus has to
1241 * rely on the futex_wake() code removing us from hash when it
1245 /* add_wait_queue is the barrier after __set_current_state. */
1246 __set_current_state(TASK_INTERRUPTIBLE);
1247 add_wait_queue(&q.waiter, &wait);
1249 * !plist_node_empty() is safe here without any lock.
1250 * q.lock_ptr != 0 is not safe, because of ordering against wakeup.
1252 if (likely(!plist_node_empty(&q.list))) {
1256 hrtimer_init_on_stack(&t.timer,
1257 clockrt ? CLOCK_REALTIME :
1260 hrtimer_init_sleeper(&t, current);
1261 hrtimer_set_expires_range_ns(&t.timer, *abs_time,
1262 current->timer_slack_ns);
1264 hrtimer_start_expires(&t.timer, HRTIMER_MODE_ABS);
1265 if (!hrtimer_active(&t.timer))
1269 * the timer could have already expired, in which
1270 * case current would be flagged for rescheduling.
1271 * Don't bother calling schedule.
1276 hrtimer_cancel(&t.timer);
1278 /* Flag if a timeout occured */
1279 rem = (t.task == NULL);
1281 destroy_hrtimer_on_stack(&t.timer);
1284 __set_current_state(TASK_RUNNING);
1287 * NOTE: we don't remove ourselves from the waitqueue because
1288 * we are the only user of it.
1291 /* If we were woken (and unqueued), we succeeded, whatever. */
1293 if (!unqueue_me(&q))
1300 * We expect signal_pending(current), but another thread may
1301 * have handled it for us already.
1307 restart = ¤t_thread_info()->restart_block;
1308 restart->fn = futex_wait_restart;
1309 restart->futex.uaddr = (u32 *)uaddr;
1310 restart->futex.val = val;
1311 restart->futex.time = abs_time->tv64;
1312 restart->futex.bitset = bitset;
1313 restart->futex.flags = 0;
1316 restart->futex.flags |= FLAGS_SHARED;
1318 restart->futex.flags |= FLAGS_CLOCKRT;
1320 ret = -ERESTART_RESTARTBLOCK;
1323 put_futex_key(fshared, &q.key);
1329 static long futex_wait_restart(struct restart_block *restart)
1331 u32 __user *uaddr = (u32 __user *)restart->futex.uaddr;
1335 t.tv64 = restart->futex.time;
1336 restart->fn = do_no_restart_syscall;
1337 if (restart->futex.flags & FLAGS_SHARED)
1339 return (long)futex_wait(uaddr, fshared, restart->futex.val, &t,
1340 restart->futex.bitset,
1341 restart->futex.flags & FLAGS_CLOCKRT);
1346 * Userspace tried a 0 -> TID atomic transition of the futex value
1347 * and failed. The kernel side here does the whole locking operation:
1348 * if there are waiters then it will block, it does PI, etc. (Due to
1349 * races the kernel might see a 0 value of the futex too.)
1351 static int futex_lock_pi(u32 __user *uaddr, int fshared,
1352 int detect, ktime_t *time, int trylock)
1354 struct hrtimer_sleeper timeout, *to = NULL;
1355 struct task_struct *curr = current;
1356 struct futex_hash_bucket *hb;
1357 u32 uval, newval, curval;
1359 int ret, lock_taken, ownerdied = 0, attempt = 0;
1361 if (refill_pi_state_cache())
1366 hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME,
1368 hrtimer_init_sleeper(to, current);
1369 hrtimer_set_expires(&to->timer, *time);
1374 q.key = FUTEX_KEY_INIT;
1375 ret = get_futex_key(uaddr, fshared, &q.key);
1376 if (unlikely(ret != 0))
1380 hb = queue_lock(&q);
1383 ret = lock_taken = 0;
1386 * To avoid races, we attempt to take the lock here again
1387 * (by doing a 0 -> TID atomic cmpxchg), while holding all
1388 * the locks. It will most likely not succeed.
1390 newval = task_pid_vnr(current);
1392 curval = cmpxchg_futex_value_locked(uaddr, 0, newval);
1394 if (unlikely(curval == -EFAULT))
1398 * Detect deadlocks. In case of REQUEUE_PI this is a valid
1399 * situation and we return success to user space.
1401 if (unlikely((curval & FUTEX_TID_MASK) == task_pid_vnr(current))) {
1403 goto out_unlock_put_key;
1407 * Surprise - we got the lock. Just return to userspace:
1409 if (unlikely(!curval))
1410 goto out_unlock_put_key;
1415 * Set the WAITERS flag, so the owner will know it has someone
1416 * to wake at next unlock
1418 newval = curval | FUTEX_WAITERS;
1421 * There are two cases, where a futex might have no owner (the
1422 * owner TID is 0): OWNER_DIED. We take over the futex in this
1423 * case. We also do an unconditional take over, when the owner
1424 * of the futex died.
1426 * This is safe as we are protected by the hash bucket lock !
1428 if (unlikely(ownerdied || !(curval & FUTEX_TID_MASK))) {
1429 /* Keep the OWNER_DIED bit */
1430 newval = (curval & ~FUTEX_TID_MASK) | task_pid_vnr(current);
1435 curval = cmpxchg_futex_value_locked(uaddr, uval, newval);
1437 if (unlikely(curval == -EFAULT))
1439 if (unlikely(curval != uval))
1443 * We took the lock due to owner died take over.
1445 if (unlikely(lock_taken))
1446 goto out_unlock_put_key;
1449 * We dont have the lock. Look up the PI state (or create it if
1450 * we are the first waiter):
1452 ret = lookup_pi_state(uval, hb, &q.key, &q.pi_state);
1454 if (unlikely(ret)) {
1459 * Task is exiting and we just wait for the
1462 queue_unlock(&q, hb);
1463 put_futex_key(fshared, &q.key);
1469 * No owner found for this futex. Check if the
1470 * OWNER_DIED bit is set to figure out whether
1471 * this is a robust futex or not.
1473 if (get_futex_value_locked(&curval, uaddr))
1477 * We simply start over in case of a robust
1478 * futex. The code above will take the futex
1481 if (curval & FUTEX_OWNER_DIED) {
1486 goto out_unlock_put_key;
1491 * Only actually queue now that the atomic ops are done:
1495 WARN_ON(!q.pi_state);
1497 * Block on the PI mutex:
1500 ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1);
1502 ret = rt_mutex_trylock(&q.pi_state->pi_mutex);
1503 /* Fixup the trylock return value: */
1504 ret = ret ? 0 : -EWOULDBLOCK;
1507 spin_lock(q.lock_ptr);
1511 * Got the lock. We might not be the anticipated owner
1512 * if we did a lock-steal - fix up the PI-state in
1515 if (q.pi_state->owner != curr)
1516 ret = fixup_pi_state_owner(uaddr, &q, curr, fshared);
1519 * Catch the rare case, where the lock was released
1520 * when we were on the way back before we locked the
1523 if (q.pi_state->owner == curr) {
1525 * Try to get the rt_mutex now. This might
1526 * fail as some other task acquired the
1527 * rt_mutex after we removed ourself from the
1528 * rt_mutex waiters list.
1530 if (rt_mutex_trylock(&q.pi_state->pi_mutex))
1534 * pi_state is incorrect, some other
1535 * task did a lock steal and we
1536 * returned due to timeout or signal
1537 * without taking the rt_mutex. Too
1538 * late. We can access the
1539 * rt_mutex_owner without locking, as
1540 * the other task is now blocked on
1541 * the hash bucket lock. Fix the state
1544 struct task_struct *owner;
1547 owner = rt_mutex_owner(&q.pi_state->pi_mutex);
1548 res = fixup_pi_state_owner(uaddr, &q, owner,
1551 /* propagate -EFAULT, if the fixup failed */
1557 * Paranoia check. If we did not take the lock
1558 * in the trylock above, then we should not be
1559 * the owner of the rtmutex, neither the real
1560 * nor the pending one:
1562 if (rt_mutex_owner(&q.pi_state->pi_mutex) == curr)
1563 printk(KERN_ERR "futex_lock_pi: ret = %d "
1564 "pi-mutex: %p pi-state %p\n", ret,
1565 q.pi_state->pi_mutex.owner,
1570 /* Unqueue and drop the lock */
1574 destroy_hrtimer_on_stack(&to->timer);
1575 return ret != -EINTR ? ret : -ERESTARTNOINTR;
1578 queue_unlock(&q, hb);
1581 put_futex_key(fshared, &q.key);
1584 destroy_hrtimer_on_stack(&to->timer);
1589 * We have to r/w *(int __user *)uaddr, and we have to modify it
1590 * atomically. Therefore, if we continue to fault after get_user()
1591 * below, we need to handle the fault ourselves, while still holding
1592 * the mmap_sem. This can occur if the uaddr is under contention as
1593 * we have to drop the mmap_sem in order to call get_user().
1595 queue_unlock(&q, hb);
1598 ret = futex_handle_fault((unsigned long)uaddr, attempt);
1601 goto retry_unlocked;
1604 ret = get_user(uval, uaddr);
1606 goto retry_unlocked;
1613 * Userspace attempted a TID -> 0 atomic transition, and failed.
1614 * This is the in-kernel slowpath: we look up the PI state (if any),
1615 * and do the rt-mutex unlock.
1617 static int futex_unlock_pi(u32 __user *uaddr, int fshared)
1619 struct futex_hash_bucket *hb;
1620 struct futex_q *this, *next;
1622 struct plist_head *head;
1623 union futex_key key = FUTEX_KEY_INIT;
1624 int ret, attempt = 0;
1627 if (get_user(uval, uaddr))
1630 * We release only a lock we actually own:
1632 if ((uval & FUTEX_TID_MASK) != task_pid_vnr(current))
1635 ret = get_futex_key(uaddr, fshared, &key);
1636 if (unlikely(ret != 0))
1639 hb = hash_futex(&key);
1641 spin_lock(&hb->lock);
1644 * To avoid races, try to do the TID -> 0 atomic transition
1645 * again. If it succeeds then we can return without waking
1648 if (!(uval & FUTEX_OWNER_DIED))
1649 uval = cmpxchg_futex_value_locked(uaddr, task_pid_vnr(current), 0);
1652 if (unlikely(uval == -EFAULT))
1655 * Rare case: we managed to release the lock atomically,
1656 * no need to wake anyone else up:
1658 if (unlikely(uval == task_pid_vnr(current)))
1662 * Ok, other tasks may need to be woken up - check waiters
1663 * and do the wakeup if necessary:
1667 plist_for_each_entry_safe(this, next, head, list) {
1668 if (!match_futex (&this->key, &key))
1670 ret = wake_futex_pi(uaddr, uval, this);
1672 * The atomic access to the futex value
1673 * generated a pagefault, so retry the
1674 * user-access and the wakeup:
1681 * No waiters - kernel unlocks the futex:
1683 if (!(uval & FUTEX_OWNER_DIED)) {
1684 ret = unlock_futex_pi(uaddr, uval);
1690 spin_unlock(&hb->lock);
1691 put_futex_key(fshared, &key);
1698 * We have to r/w *(int __user *)uaddr, and we have to modify it
1699 * atomically. Therefore, if we continue to fault after get_user()
1700 * below, we need to handle the fault ourselves, while still holding
1701 * the mmap_sem. This can occur if the uaddr is under contention as
1702 * we have to drop the mmap_sem in order to call get_user().
1704 spin_unlock(&hb->lock);
1707 ret = futex_handle_fault((unsigned long)uaddr, attempt);
1711 goto retry_unlocked;
1714 ret = get_user(uval, uaddr);
1715 put_futex_key(fshared, &key);
1723 * Support for robust futexes: the kernel cleans up held futexes at
1726 * Implementation: user-space maintains a per-thread list of locks it
1727 * is holding. Upon do_exit(), the kernel carefully walks this list,
1728 * and marks all locks that are owned by this thread with the
1729 * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is
1730 * always manipulated with the lock held, so the list is private and
1731 * per-thread. Userspace also maintains a per-thread 'list_op_pending'
1732 * field, to allow the kernel to clean up if the thread dies after
1733 * acquiring the lock, but just before it could have added itself to
1734 * the list. There can only be one such pending lock.
1738 * sys_set_robust_list - set the robust-futex list head of a task
1739 * @head: pointer to the list-head
1740 * @len: length of the list-head, as userspace expects
1742 SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head,
1745 if (!futex_cmpxchg_enabled)
1748 * The kernel knows only one size for now:
1750 if (unlikely(len != sizeof(*head)))
1753 current->robust_list = head;
1759 * sys_get_robust_list - get the robust-futex list head of a task
1760 * @pid: pid of the process [zero for current task]
1761 * @head_ptr: pointer to a list-head pointer, the kernel fills it in
1762 * @len_ptr: pointer to a length field, the kernel fills in the header size
1764 SYSCALL_DEFINE3(get_robust_list, int, pid,
1765 struct robust_list_head __user * __user *, head_ptr,
1766 size_t __user *, len_ptr)
1768 struct robust_list_head __user *head;
1770 const struct cred *cred = current_cred(), *pcred;
1772 if (!futex_cmpxchg_enabled)
1776 head = current->robust_list;
1778 struct task_struct *p;
1782 p = find_task_by_vpid(pid);
1786 pcred = __task_cred(p);
1787 if (cred->euid != pcred->euid &&
1788 cred->euid != pcred->uid &&
1789 !capable(CAP_SYS_PTRACE))
1791 head = p->robust_list;
1795 if (put_user(sizeof(*head), len_ptr))
1797 return put_user(head, head_ptr);
1806 * Process a futex-list entry, check whether it's owned by the
1807 * dying task, and do notification if so:
1809 int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi)
1811 u32 uval, nval, mval;
1814 if (get_user(uval, uaddr))
1817 if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) {
1819 * Ok, this dying thread is truly holding a futex
1820 * of interest. Set the OWNER_DIED bit atomically
1821 * via cmpxchg, and if the value had FUTEX_WAITERS
1822 * set, wake up a waiter (if any). (We have to do a
1823 * futex_wake() even if OWNER_DIED is already set -
1824 * to handle the rare but possible case of recursive
1825 * thread-death.) The rest of the cleanup is done in
1828 mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED;
1829 nval = futex_atomic_cmpxchg_inatomic(uaddr, uval, mval);
1831 if (nval == -EFAULT)
1838 * Wake robust non-PI futexes here. The wakeup of
1839 * PI futexes happens in exit_pi_state():
1841 if (!pi && (uval & FUTEX_WAITERS))
1842 futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY);
1848 * Fetch a robust-list pointer. Bit 0 signals PI futexes:
1850 static inline int fetch_robust_entry(struct robust_list __user **entry,
1851 struct robust_list __user * __user *head,
1854 unsigned long uentry;
1856 if (get_user(uentry, (unsigned long __user *)head))
1859 *entry = (void __user *)(uentry & ~1UL);
1866 * Walk curr->robust_list (very carefully, it's a userspace list!)
1867 * and mark any locks found there dead, and notify any waiters.
1869 * We silently return on any sign of list-walking problem.
1871 void exit_robust_list(struct task_struct *curr)
1873 struct robust_list_head __user *head = curr->robust_list;
1874 struct robust_list __user *entry, *next_entry, *pending;
1875 unsigned int limit = ROBUST_LIST_LIMIT, pi, next_pi, pip;
1876 unsigned long futex_offset;
1879 if (!futex_cmpxchg_enabled)
1883 * Fetch the list head (which was registered earlier, via
1884 * sys_set_robust_list()):
1886 if (fetch_robust_entry(&entry, &head->list.next, &pi))
1889 * Fetch the relative futex offset:
1891 if (get_user(futex_offset, &head->futex_offset))
1894 * Fetch any possibly pending lock-add first, and handle it
1897 if (fetch_robust_entry(&pending, &head->list_op_pending, &pip))
1900 next_entry = NULL; /* avoid warning with gcc */
1901 while (entry != &head->list) {
1903 * Fetch the next entry in the list before calling
1904 * handle_futex_death:
1906 rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi);
1908 * A pending lock might already be on the list, so
1909 * don't process it twice:
1911 if (entry != pending)
1912 if (handle_futex_death((void __user *)entry + futex_offset,
1920 * Avoid excessively long or circular lists:
1929 handle_futex_death((void __user *)pending + futex_offset,
1933 long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout,
1934 u32 __user *uaddr2, u32 val2, u32 val3)
1936 int clockrt, ret = -ENOSYS;
1937 int cmd = op & FUTEX_CMD_MASK;
1940 if (!(op & FUTEX_PRIVATE_FLAG))
1943 clockrt = op & FUTEX_CLOCK_REALTIME;
1944 if (clockrt && cmd != FUTEX_WAIT_BITSET)
1949 val3 = FUTEX_BITSET_MATCH_ANY;
1950 case FUTEX_WAIT_BITSET:
1951 ret = futex_wait(uaddr, fshared, val, timeout, val3, clockrt);
1954 val3 = FUTEX_BITSET_MATCH_ANY;
1955 case FUTEX_WAKE_BITSET:
1956 ret = futex_wake(uaddr, fshared, val, val3);
1959 ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, NULL);
1961 case FUTEX_CMP_REQUEUE:
1962 ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, &val3);
1965 ret = futex_wake_op(uaddr, fshared, uaddr2, val, val2, val3);
1968 if (futex_cmpxchg_enabled)
1969 ret = futex_lock_pi(uaddr, fshared, val, timeout, 0);
1971 case FUTEX_UNLOCK_PI:
1972 if (futex_cmpxchg_enabled)
1973 ret = futex_unlock_pi(uaddr, fshared);
1975 case FUTEX_TRYLOCK_PI:
1976 if (futex_cmpxchg_enabled)
1977 ret = futex_lock_pi(uaddr, fshared, 0, timeout, 1);
1986 SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val,
1987 struct timespec __user *, utime, u32 __user *, uaddr2,
1991 ktime_t t, *tp = NULL;
1993 int cmd = op & FUTEX_CMD_MASK;
1995 if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI ||
1996 cmd == FUTEX_WAIT_BITSET)) {
1997 if (copy_from_user(&ts, utime, sizeof(ts)) != 0)
1999 if (!timespec_valid(&ts))
2002 t = timespec_to_ktime(ts);
2003 if (cmd == FUTEX_WAIT)
2004 t = ktime_add_safe(ktime_get(), t);
2008 * requeue parameter in 'utime' if cmd == FUTEX_REQUEUE.
2009 * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP.
2011 if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE ||
2012 cmd == FUTEX_WAKE_OP)
2013 val2 = (u32) (unsigned long) utime;
2015 return do_futex(uaddr, op, val, tp, uaddr2, val2, val3);
2018 static int __init futex_init(void)
2024 * This will fail and we want it. Some arch implementations do
2025 * runtime detection of the futex_atomic_cmpxchg_inatomic()
2026 * functionality. We want to know that before we call in any
2027 * of the complex code paths. Also we want to prevent
2028 * registration of robust lists in that case. NULL is
2029 * guaranteed to fault and we get -EFAULT on functional
2030 * implementation, the non functional ones will return
2033 curval = cmpxchg_futex_value_locked(NULL, 0, 0);
2034 if (curval == -EFAULT)
2035 futex_cmpxchg_enabled = 1;
2037 for (i = 0; i < ARRAY_SIZE(futex_queues); i++) {
2038 plist_head_init(&futex_queues[i].chain, &futex_queues[i].lock);
2039 spin_lock_init(&futex_queues[i].lock);
2044 __initcall(futex_init);