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
)
176 switch (key
->both
.offset
& (FUT_OFF_INODE
|FUT_OFF_MMSHARED
)) {
178 iput(key
->shared
.inode
);
180 case FUT_OFF_MMSHARED
:
181 mmdrop(key
->private.mm
);
187 * get_futex_key - Get parameters which are the keys for a futex.
188 * @uaddr: virtual address of the futex
189 * @shared: NULL for a PROCESS_PRIVATE futex,
190 * ¤t->mm->mmap_sem for a PROCESS_SHARED futex
191 * @key: address where result is stored.
193 * Returns a negative error code or 0
194 * The key words are stored in *key on success.
196 * For shared mappings, it's (page->index, vma->vm_file->f_path.dentry->d_inode,
197 * offset_within_page). For private mappings, it's (uaddr, current->mm).
198 * We can usually work out the index without swapping in the page.
200 * fshared is NULL for PROCESS_PRIVATE futexes
201 * For other futexes, it points to ¤t->mm->mmap_sem and
202 * caller must have taken the reader lock. but NOT any spinlocks.
204 static int get_futex_key(u32 __user
*uaddr
, int fshared
, 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
;
231 get_futex_key_refs(key
);
236 err
= get_user_pages_fast(address
, 1, 0, &page
);
241 if (!page
->mapping
) {
248 * Private mappings are handled in a simple way.
250 * NOTE: When userspace waits on a MAP_SHARED mapping, even if
251 * it's a read-only handle, it's expected that futexes attach to
252 * the object not the particular process.
254 if (PageAnon(page
)) {
255 key
->both
.offset
|= FUT_OFF_MMSHARED
; /* ref taken on mm */
256 key
->private.mm
= mm
;
257 key
->private.address
= address
;
259 key
->both
.offset
|= FUT_OFF_INODE
; /* inode-based key */
260 key
->shared
.inode
= page
->mapping
->host
;
261 key
->shared
.pgoff
= page
->index
;
264 get_futex_key_refs(key
);
272 void put_futex_key(int fshared
, union futex_key
*key
)
274 drop_futex_key_refs(key
);
277 static u32
cmpxchg_futex_value_locked(u32 __user
*uaddr
, u32 uval
, u32 newval
)
282 curval
= futex_atomic_cmpxchg_inatomic(uaddr
, uval
, newval
);
288 static int get_futex_value_locked(u32
*dest
, u32 __user
*from
)
293 ret
= __copy_from_user_inatomic(dest
, from
, sizeof(u32
));
296 return ret
? -EFAULT
: 0;
302 static int futex_handle_fault(unsigned long address
, int attempt
)
304 struct vm_area_struct
* vma
;
305 struct mm_struct
*mm
= current
->mm
;
311 down_read(&mm
->mmap_sem
);
312 vma
= find_vma(mm
, address
);
313 if (vma
&& address
>= vma
->vm_start
&&
314 (vma
->vm_flags
& VM_WRITE
)) {
316 fault
= handle_mm_fault(mm
, vma
, address
, 1);
317 if (unlikely((fault
& VM_FAULT_ERROR
))) {
319 /* XXX: let's do this when we verify it is OK */
320 if (ret
& VM_FAULT_OOM
)
325 if (fault
& VM_FAULT_MAJOR
)
331 up_read(&mm
->mmap_sem
);
338 static int refill_pi_state_cache(void)
340 struct futex_pi_state
*pi_state
;
342 if (likely(current
->pi_state_cache
))
345 pi_state
= kzalloc(sizeof(*pi_state
), GFP_KERNEL
);
350 INIT_LIST_HEAD(&pi_state
->list
);
351 /* pi_mutex gets initialized later */
352 pi_state
->owner
= NULL
;
353 atomic_set(&pi_state
->refcount
, 1);
354 pi_state
->key
= FUTEX_KEY_INIT
;
356 current
->pi_state_cache
= pi_state
;
361 static struct futex_pi_state
* alloc_pi_state(void)
363 struct futex_pi_state
*pi_state
= current
->pi_state_cache
;
366 current
->pi_state_cache
= NULL
;
371 static void free_pi_state(struct futex_pi_state
*pi_state
)
373 if (!atomic_dec_and_test(&pi_state
->refcount
))
377 * If pi_state->owner is NULL, the owner is most probably dying
378 * and has cleaned up the pi_state already
380 if (pi_state
->owner
) {
381 spin_lock_irq(&pi_state
->owner
->pi_lock
);
382 list_del_init(&pi_state
->list
);
383 spin_unlock_irq(&pi_state
->owner
->pi_lock
);
385 rt_mutex_proxy_unlock(&pi_state
->pi_mutex
, pi_state
->owner
);
388 if (current
->pi_state_cache
)
392 * pi_state->list is already empty.
393 * clear pi_state->owner.
394 * refcount is at 0 - put it back to 1.
396 pi_state
->owner
= NULL
;
397 atomic_set(&pi_state
->refcount
, 1);
398 current
->pi_state_cache
= pi_state
;
403 * Look up the task based on what TID userspace gave us.
406 static struct task_struct
* futex_find_get_task(pid_t pid
)
408 struct task_struct
*p
;
411 p
= find_task_by_vpid(pid
);
412 if (!p
|| ((current
->euid
!= p
->euid
) && (current
->euid
!= p
->uid
)))
423 * This task is holding PI mutexes at exit time => bad.
424 * Kernel cleans up PI-state, but userspace is likely hosed.
425 * (Robust-futex cleanup is separate and might save the day for userspace.)
427 void exit_pi_state_list(struct task_struct
*curr
)
429 struct list_head
*next
, *head
= &curr
->pi_state_list
;
430 struct futex_pi_state
*pi_state
;
431 struct futex_hash_bucket
*hb
;
432 union futex_key key
= FUTEX_KEY_INIT
;
434 if (!futex_cmpxchg_enabled
)
437 * We are a ZOMBIE and nobody can enqueue itself on
438 * pi_state_list anymore, but we have to be careful
439 * versus waiters unqueueing themselves:
441 spin_lock_irq(&curr
->pi_lock
);
442 while (!list_empty(head
)) {
445 pi_state
= list_entry(next
, struct futex_pi_state
, list
);
447 hb
= hash_futex(&key
);
448 spin_unlock_irq(&curr
->pi_lock
);
450 spin_lock(&hb
->lock
);
452 spin_lock_irq(&curr
->pi_lock
);
454 * We dropped the pi-lock, so re-check whether this
455 * task still owns the PI-state:
457 if (head
->next
!= next
) {
458 spin_unlock(&hb
->lock
);
462 WARN_ON(pi_state
->owner
!= curr
);
463 WARN_ON(list_empty(&pi_state
->list
));
464 list_del_init(&pi_state
->list
);
465 pi_state
->owner
= NULL
;
466 spin_unlock_irq(&curr
->pi_lock
);
468 rt_mutex_unlock(&pi_state
->pi_mutex
);
470 spin_unlock(&hb
->lock
);
472 spin_lock_irq(&curr
->pi_lock
);
474 spin_unlock_irq(&curr
->pi_lock
);
478 lookup_pi_state(u32 uval
, struct futex_hash_bucket
*hb
,
479 union futex_key
*key
, struct futex_pi_state
**ps
)
481 struct futex_pi_state
*pi_state
= NULL
;
482 struct futex_q
*this, *next
;
483 struct plist_head
*head
;
484 struct task_struct
*p
;
485 pid_t pid
= uval
& FUTEX_TID_MASK
;
489 plist_for_each_entry_safe(this, next
, head
, list
) {
490 if (match_futex(&this->key
, key
)) {
492 * Another waiter already exists - bump up
493 * the refcount and return its pi_state:
495 pi_state
= this->pi_state
;
497 * Userspace might have messed up non PI and PI futexes
499 if (unlikely(!pi_state
))
502 WARN_ON(!atomic_read(&pi_state
->refcount
));
503 WARN_ON(pid
&& pi_state
->owner
&&
504 pi_state
->owner
->pid
!= pid
);
506 atomic_inc(&pi_state
->refcount
);
514 * We are the first waiter - try to look up the real owner and attach
515 * the new pi_state to it, but bail out when TID = 0
519 p
= futex_find_get_task(pid
);
524 * We need to look at the task state flags to figure out,
525 * whether the task is exiting. To protect against the do_exit
526 * change of the task flags, we do this protected by
529 spin_lock_irq(&p
->pi_lock
);
530 if (unlikely(p
->flags
& PF_EXITING
)) {
532 * The task is on the way out. When PF_EXITPIDONE is
533 * set, we know that the task has finished the
536 int ret
= (p
->flags
& PF_EXITPIDONE
) ? -ESRCH
: -EAGAIN
;
538 spin_unlock_irq(&p
->pi_lock
);
543 pi_state
= alloc_pi_state();
546 * Initialize the pi_mutex in locked state and make 'p'
549 rt_mutex_init_proxy_locked(&pi_state
->pi_mutex
, p
);
551 /* Store the key for possible exit cleanups: */
552 pi_state
->key
= *key
;
554 WARN_ON(!list_empty(&pi_state
->list
));
555 list_add(&pi_state
->list
, &p
->pi_state_list
);
557 spin_unlock_irq(&p
->pi_lock
);
567 * The hash bucket lock must be held when this is called.
568 * Afterwards, the futex_q must not be accessed.
570 static void wake_futex(struct futex_q
*q
)
572 plist_del(&q
->list
, &q
->list
.plist
);
574 * The lock in wake_up_all() is a crucial memory barrier after the
575 * plist_del() and also before assigning to q->lock_ptr.
579 * The waiting task can free the futex_q as soon as this is written,
580 * without taking any locks. This must come last.
582 * A memory barrier is required here to prevent the following store
583 * to lock_ptr from getting ahead of the wakeup. Clearing the lock
584 * at the end of wake_up_all() does not prevent this store from
591 static int wake_futex_pi(u32 __user
*uaddr
, u32 uval
, struct futex_q
*this)
593 struct task_struct
*new_owner
;
594 struct futex_pi_state
*pi_state
= this->pi_state
;
600 spin_lock(&pi_state
->pi_mutex
.wait_lock
);
601 new_owner
= rt_mutex_next_owner(&pi_state
->pi_mutex
);
604 * This happens when we have stolen the lock and the original
605 * pending owner did not enqueue itself back on the rt_mutex.
606 * Thats not a tragedy. We know that way, that a lock waiter
607 * is on the fly. We make the futex_q waiter the pending owner.
610 new_owner
= this->task
;
613 * We pass it to the next owner. (The WAITERS bit is always
614 * kept enabled while there is PI state around. We must also
615 * preserve the owner died bit.)
617 if (!(uval
& FUTEX_OWNER_DIED
)) {
620 newval
= FUTEX_WAITERS
| task_pid_vnr(new_owner
);
622 curval
= cmpxchg_futex_value_locked(uaddr
, uval
, newval
);
624 if (curval
== -EFAULT
)
626 else if (curval
!= uval
)
629 spin_unlock(&pi_state
->pi_mutex
.wait_lock
);
634 spin_lock_irq(&pi_state
->owner
->pi_lock
);
635 WARN_ON(list_empty(&pi_state
->list
));
636 list_del_init(&pi_state
->list
);
637 spin_unlock_irq(&pi_state
->owner
->pi_lock
);
639 spin_lock_irq(&new_owner
->pi_lock
);
640 WARN_ON(!list_empty(&pi_state
->list
));
641 list_add(&pi_state
->list
, &new_owner
->pi_state_list
);
642 pi_state
->owner
= new_owner
;
643 spin_unlock_irq(&new_owner
->pi_lock
);
645 spin_unlock(&pi_state
->pi_mutex
.wait_lock
);
646 rt_mutex_unlock(&pi_state
->pi_mutex
);
651 static int unlock_futex_pi(u32 __user
*uaddr
, u32 uval
)
656 * There is no waiter, so we unlock the futex. The owner died
657 * bit has not to be preserved here. We are the owner:
659 oldval
= cmpxchg_futex_value_locked(uaddr
, uval
, 0);
661 if (oldval
== -EFAULT
)
670 * Express the locking dependencies for lockdep:
673 double_lock_hb(struct futex_hash_bucket
*hb1
, struct futex_hash_bucket
*hb2
)
676 spin_lock(&hb1
->lock
);
678 spin_lock_nested(&hb2
->lock
, SINGLE_DEPTH_NESTING
);
679 } else { /* hb1 > hb2 */
680 spin_lock(&hb2
->lock
);
681 spin_lock_nested(&hb1
->lock
, SINGLE_DEPTH_NESTING
);
686 * Wake up all waiters hashed on the physical page that is mapped
687 * to this virtual address:
689 static int futex_wake(u32 __user
*uaddr
, int fshared
, int nr_wake
, u32 bitset
)
691 struct futex_hash_bucket
*hb
;
692 struct futex_q
*this, *next
;
693 struct plist_head
*head
;
694 union futex_key key
= FUTEX_KEY_INIT
;
700 ret
= get_futex_key(uaddr
, fshared
, &key
);
701 if (unlikely(ret
!= 0))
704 hb
= hash_futex(&key
);
705 spin_lock(&hb
->lock
);
708 plist_for_each_entry_safe(this, next
, head
, list
) {
709 if (match_futex (&this->key
, &key
)) {
710 if (this->pi_state
) {
715 /* Check if one of the bits is set in both bitsets */
716 if (!(this->bitset
& bitset
))
720 if (++ret
>= nr_wake
)
725 spin_unlock(&hb
->lock
);
727 put_futex_key(fshared
, &key
);
732 * Wake up all waiters hashed on the physical page that is mapped
733 * to this virtual address:
736 futex_wake_op(u32 __user
*uaddr1
, int fshared
, u32 __user
*uaddr2
,
737 int nr_wake
, int nr_wake2
, int op
)
739 union futex_key key1
= FUTEX_KEY_INIT
, key2
= FUTEX_KEY_INIT
;
740 struct futex_hash_bucket
*hb1
, *hb2
;
741 struct plist_head
*head
;
742 struct futex_q
*this, *next
;
743 int ret
, op_ret
, attempt
= 0;
746 ret
= get_futex_key(uaddr1
, fshared
, &key1
);
747 if (unlikely(ret
!= 0))
749 ret
= get_futex_key(uaddr2
, fshared
, &key2
);
750 if (unlikely(ret
!= 0))
753 hb1
= hash_futex(&key1
);
754 hb2
= hash_futex(&key2
);
757 double_lock_hb(hb1
, hb2
);
759 op_ret
= futex_atomic_op_inuser(op
, uaddr2
);
760 if (unlikely(op_ret
< 0)) {
763 spin_unlock(&hb1
->lock
);
765 spin_unlock(&hb2
->lock
);
769 * we don't get EFAULT from MMU faults if we don't have an MMU,
770 * but we might get them from range checking
776 if (unlikely(op_ret
!= -EFAULT
)) {
782 * futex_atomic_op_inuser needs to both read and write
783 * *(int __user *)uaddr2, but we can't modify it
784 * non-atomically. Therefore, if get_user below is not
785 * enough, we need to handle the fault ourselves, while
786 * still holding the mmap_sem.
789 ret
= futex_handle_fault((unsigned long)uaddr2
,
796 ret
= get_user(dummy
, uaddr2
);
805 plist_for_each_entry_safe(this, next
, head
, list
) {
806 if (match_futex (&this->key
, &key1
)) {
808 if (++ret
>= nr_wake
)
817 plist_for_each_entry_safe(this, next
, head
, list
) {
818 if (match_futex (&this->key
, &key2
)) {
820 if (++op_ret
>= nr_wake2
)
827 spin_unlock(&hb1
->lock
);
829 spin_unlock(&hb2
->lock
);
831 put_futex_key(fshared
, &key2
);
832 put_futex_key(fshared
, &key1
);
838 * Requeue all waiters hashed on one physical page to another
841 static int futex_requeue(u32 __user
*uaddr1
, int fshared
, u32 __user
*uaddr2
,
842 int nr_wake
, int nr_requeue
, u32
*cmpval
)
844 union futex_key key1
= FUTEX_KEY_INIT
, key2
= FUTEX_KEY_INIT
;
845 struct futex_hash_bucket
*hb1
, *hb2
;
846 struct plist_head
*head1
;
847 struct futex_q
*this, *next
;
848 int ret
, drop_count
= 0;
851 ret
= get_futex_key(uaddr1
, fshared
, &key1
);
852 if (unlikely(ret
!= 0))
854 ret
= get_futex_key(uaddr2
, fshared
, &key2
);
855 if (unlikely(ret
!= 0))
858 hb1
= hash_futex(&key1
);
859 hb2
= hash_futex(&key2
);
861 double_lock_hb(hb1
, hb2
);
863 if (likely(cmpval
!= NULL
)) {
866 ret
= get_futex_value_locked(&curval
, uaddr1
);
869 spin_unlock(&hb1
->lock
);
871 spin_unlock(&hb2
->lock
);
873 ret
= get_user(curval
, uaddr1
);
880 if (curval
!= *cmpval
) {
887 plist_for_each_entry_safe(this, next
, head1
, list
) {
888 if (!match_futex (&this->key
, &key1
))
890 if (++ret
<= nr_wake
) {
894 * If key1 and key2 hash to the same bucket, no need to
897 if (likely(head1
!= &hb2
->chain
)) {
898 plist_del(&this->list
, &hb1
->chain
);
899 plist_add(&this->list
, &hb2
->chain
);
900 this->lock_ptr
= &hb2
->lock
;
901 #ifdef CONFIG_DEBUG_PI_LIST
902 this->list
.plist
.lock
= &hb2
->lock
;
906 get_futex_key_refs(&key2
);
909 if (ret
- nr_wake
>= nr_requeue
)
915 spin_unlock(&hb1
->lock
);
917 spin_unlock(&hb2
->lock
);
919 /* drop_futex_key_refs() must be called outside the spinlocks. */
920 while (--drop_count
>= 0)
921 drop_futex_key_refs(&key1
);
924 put_futex_key(fshared
, &key2
);
925 put_futex_key(fshared
, &key1
);
929 /* The key must be already stored in q->key. */
930 static inline struct futex_hash_bucket
*queue_lock(struct futex_q
*q
)
932 struct futex_hash_bucket
*hb
;
934 init_waitqueue_head(&q
->waiter
);
936 get_futex_key_refs(&q
->key
);
937 hb
= hash_futex(&q
->key
);
938 q
->lock_ptr
= &hb
->lock
;
940 spin_lock(&hb
->lock
);
944 static inline void queue_me(struct futex_q
*q
, struct futex_hash_bucket
*hb
)
949 * The priority used to register this element is
950 * - either the real thread-priority for the real-time threads
951 * (i.e. threads with a priority lower than MAX_RT_PRIO)
952 * - or MAX_RT_PRIO for non-RT threads.
953 * Thus, all RT-threads are woken first in priority order, and
954 * the others are woken last, in FIFO order.
956 prio
= min(current
->normal_prio
, MAX_RT_PRIO
);
958 plist_node_init(&q
->list
, prio
);
959 #ifdef CONFIG_DEBUG_PI_LIST
960 q
->list
.plist
.lock
= &hb
->lock
;
962 plist_add(&q
->list
, &hb
->chain
);
964 spin_unlock(&hb
->lock
);
968 queue_unlock(struct futex_q
*q
, struct futex_hash_bucket
*hb
)
970 spin_unlock(&hb
->lock
);
971 drop_futex_key_refs(&q
->key
);
975 * queue_me and unqueue_me must be called as a pair, each
976 * exactly once. They are called with the hashed spinlock held.
979 /* Return 1 if we were still queued (ie. 0 means we were woken) */
980 static int unqueue_me(struct futex_q
*q
)
982 spinlock_t
*lock_ptr
;
985 /* In the common case we don't take the spinlock, which is nice. */
987 lock_ptr
= q
->lock_ptr
;
989 if (lock_ptr
!= NULL
) {
992 * q->lock_ptr can change between reading it and
993 * spin_lock(), causing us to take the wrong lock. This
994 * corrects the race condition.
996 * Reasoning goes like this: if we have the wrong lock,
997 * q->lock_ptr must have changed (maybe several times)
998 * between reading it and the spin_lock(). It can
999 * change again after the spin_lock() but only if it was
1000 * already changed before the spin_lock(). It cannot,
1001 * however, change back to the original value. Therefore
1002 * we can detect whether we acquired the correct lock.
1004 if (unlikely(lock_ptr
!= q
->lock_ptr
)) {
1005 spin_unlock(lock_ptr
);
1008 WARN_ON(plist_node_empty(&q
->list
));
1009 plist_del(&q
->list
, &q
->list
.plist
);
1011 BUG_ON(q
->pi_state
);
1013 spin_unlock(lock_ptr
);
1017 drop_futex_key_refs(&q
->key
);
1022 * PI futexes can not be requeued and must remove themself from the
1023 * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry
1026 static void unqueue_me_pi(struct futex_q
*q
)
1028 WARN_ON(plist_node_empty(&q
->list
));
1029 plist_del(&q
->list
, &q
->list
.plist
);
1031 BUG_ON(!q
->pi_state
);
1032 free_pi_state(q
->pi_state
);
1035 spin_unlock(q
->lock_ptr
);
1037 drop_futex_key_refs(&q
->key
);
1041 * Fixup the pi_state owner with the new owner.
1043 * Must be called with hash bucket lock held and mm->sem held for non
1046 static int fixup_pi_state_owner(u32 __user
*uaddr
, struct futex_q
*q
,
1047 struct task_struct
*newowner
, int fshared
)
1049 u32 newtid
= task_pid_vnr(newowner
) | FUTEX_WAITERS
;
1050 struct futex_pi_state
*pi_state
= q
->pi_state
;
1051 struct task_struct
*oldowner
= pi_state
->owner
;
1052 u32 uval
, curval
, newval
;
1053 int ret
, attempt
= 0;
1056 if (!pi_state
->owner
)
1057 newtid
|= FUTEX_OWNER_DIED
;
1060 * We are here either because we stole the rtmutex from the
1061 * pending owner or we are the pending owner which failed to
1062 * get the rtmutex. We have to replace the pending owner TID
1063 * in the user space variable. This must be atomic as we have
1064 * to preserve the owner died bit here.
1066 * Note: We write the user space value _before_ changing the
1067 * pi_state because we can fault here. Imagine swapped out
1068 * pages or a fork, which was running right before we acquired
1069 * mmap_sem, that marked all the anonymous memory readonly for
1072 * Modifying pi_state _before_ the user space value would
1073 * leave the pi_state in an inconsistent state when we fault
1074 * here, because we need to drop the hash bucket lock to
1075 * handle the fault. This might be observed in the PID check
1076 * in lookup_pi_state.
1079 if (get_futex_value_locked(&uval
, uaddr
))
1083 newval
= (uval
& FUTEX_OWNER_DIED
) | newtid
;
1085 curval
= cmpxchg_futex_value_locked(uaddr
, uval
, newval
);
1087 if (curval
== -EFAULT
)
1095 * We fixed up user space. Now we need to fix the pi_state
1098 if (pi_state
->owner
!= NULL
) {
1099 spin_lock_irq(&pi_state
->owner
->pi_lock
);
1100 WARN_ON(list_empty(&pi_state
->list
));
1101 list_del_init(&pi_state
->list
);
1102 spin_unlock_irq(&pi_state
->owner
->pi_lock
);
1105 pi_state
->owner
= newowner
;
1107 spin_lock_irq(&newowner
->pi_lock
);
1108 WARN_ON(!list_empty(&pi_state
->list
));
1109 list_add(&pi_state
->list
, &newowner
->pi_state_list
);
1110 spin_unlock_irq(&newowner
->pi_lock
);
1114 * To handle the page fault we need to drop the hash bucket
1115 * lock here. That gives the other task (either the pending
1116 * owner itself or the task which stole the rtmutex) the
1117 * chance to try the fixup of the pi_state. So once we are
1118 * back from handling the fault we need to check the pi_state
1119 * after reacquiring the hash bucket lock and before trying to
1120 * do another fixup. When the fixup has been done already we
1124 spin_unlock(q
->lock_ptr
);
1126 ret
= futex_handle_fault((unsigned long)uaddr
, attempt
++);
1128 spin_lock(q
->lock_ptr
);
1131 * Check if someone else fixed it for us:
1133 if (pi_state
->owner
!= oldowner
)
1143 * In case we must use restart_block to restart a futex_wait,
1144 * we encode in the 'flags' shared capability
1146 #define FLAGS_SHARED 0x01
1147 #define FLAGS_CLOCKRT 0x02
1149 static long futex_wait_restart(struct restart_block
*restart
);
1151 static int futex_wait(u32 __user
*uaddr
, int fshared
,
1152 u32 val
, ktime_t
*abs_time
, u32 bitset
, int clockrt
)
1154 struct task_struct
*curr
= current
;
1155 DECLARE_WAITQUEUE(wait
, curr
);
1156 struct futex_hash_bucket
*hb
;
1160 struct hrtimer_sleeper t
;
1169 q
.key
= FUTEX_KEY_INIT
;
1170 ret
= get_futex_key(uaddr
, fshared
, &q
.key
);
1171 if (unlikely(ret
!= 0))
1172 goto out_release_sem
;
1174 hb
= queue_lock(&q
);
1177 * Access the page AFTER the futex is queued.
1178 * Order is important:
1180 * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val);
1181 * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); }
1183 * The basic logical guarantee of a futex is that it blocks ONLY
1184 * if cond(var) is known to be true at the time of blocking, for
1185 * any cond. If we queued after testing *uaddr, that would open
1186 * a race condition where we could block indefinitely with
1187 * cond(var) false, which would violate the guarantee.
1189 * A consequence is that futex_wait() can return zero and absorb
1190 * a wakeup when *uaddr != val on entry to the syscall. This is
1193 * for shared futexes, we hold the mmap semaphore, so the mapping
1194 * cannot have changed since we looked it up in get_futex_key.
1196 ret
= get_futex_value_locked(&uval
, uaddr
);
1198 if (unlikely(ret
)) {
1199 queue_unlock(&q
, hb
);
1201 ret
= get_user(uval
, uaddr
);
1209 goto out_unlock_release_sem
;
1211 /* Only actually queue if *uaddr contained val. */
1215 * There might have been scheduling since the queue_me(), as we
1216 * cannot hold a spinlock across the get_user() in case it
1217 * faults, and we cannot just set TASK_INTERRUPTIBLE state when
1218 * queueing ourselves into the futex hash. This code thus has to
1219 * rely on the futex_wake() code removing us from hash when it
1223 /* add_wait_queue is the barrier after __set_current_state. */
1224 __set_current_state(TASK_INTERRUPTIBLE
);
1225 add_wait_queue(&q
.waiter
, &wait
);
1227 * !plist_node_empty() is safe here without any lock.
1228 * q.lock_ptr != 0 is not safe, because of ordering against wakeup.
1230 if (likely(!plist_node_empty(&q
.list
))) {
1234 unsigned long slack
;
1235 slack
= current
->timer_slack_ns
;
1236 if (rt_task(current
))
1238 hrtimer_init_on_stack(&t
.timer
,
1239 clockrt
? CLOCK_REALTIME
:
1242 hrtimer_init_sleeper(&t
, current
);
1243 hrtimer_set_expires_range_ns(&t
.timer
, *abs_time
, slack
);
1245 hrtimer_start_expires(&t
.timer
, HRTIMER_MODE_ABS
);
1246 if (!hrtimer_active(&t
.timer
))
1250 * the timer could have already expired, in which
1251 * case current would be flagged for rescheduling.
1252 * Don't bother calling schedule.
1257 hrtimer_cancel(&t
.timer
);
1259 /* Flag if a timeout occured */
1260 rem
= (t
.task
== NULL
);
1262 destroy_hrtimer_on_stack(&t
.timer
);
1265 __set_current_state(TASK_RUNNING
);
1268 * NOTE: we don't remove ourselves from the waitqueue because
1269 * we are the only user of it.
1272 /* If we were woken (and unqueued), we succeeded, whatever. */
1273 if (!unqueue_me(&q
))
1279 * We expect signal_pending(current), but another thread may
1280 * have handled it for us already.
1283 return -ERESTARTSYS
;
1285 struct restart_block
*restart
;
1286 restart
= ¤t_thread_info()->restart_block
;
1287 restart
->fn
= futex_wait_restart
;
1288 restart
->futex
.uaddr
= (u32
*)uaddr
;
1289 restart
->futex
.val
= val
;
1290 restart
->futex
.time
= abs_time
->tv64
;
1291 restart
->futex
.bitset
= bitset
;
1292 restart
->futex
.flags
= 0;
1295 restart
->futex
.flags
|= FLAGS_SHARED
;
1297 restart
->futex
.flags
|= FLAGS_CLOCKRT
;
1298 return -ERESTART_RESTARTBLOCK
;
1301 out_unlock_release_sem
:
1302 queue_unlock(&q
, hb
);
1305 put_futex_key(fshared
, &q
.key
);
1310 static long futex_wait_restart(struct restart_block
*restart
)
1312 u32 __user
*uaddr
= (u32 __user
*)restart
->futex
.uaddr
;
1316 t
.tv64
= restart
->futex
.time
;
1317 restart
->fn
= do_no_restart_syscall
;
1318 if (restart
->futex
.flags
& FLAGS_SHARED
)
1320 return (long)futex_wait(uaddr
, fshared
, restart
->futex
.val
, &t
,
1321 restart
->futex
.bitset
,
1322 restart
->futex
.flags
& FLAGS_CLOCKRT
);
1327 * Userspace tried a 0 -> TID atomic transition of the futex value
1328 * and failed. The kernel side here does the whole locking operation:
1329 * if there are waiters then it will block, it does PI, etc. (Due to
1330 * races the kernel might see a 0 value of the futex too.)
1332 static int futex_lock_pi(u32 __user
*uaddr
, int fshared
,
1333 int detect
, ktime_t
*time
, int trylock
)
1335 struct hrtimer_sleeper timeout
, *to
= NULL
;
1336 struct task_struct
*curr
= current
;
1337 struct futex_hash_bucket
*hb
;
1338 u32 uval
, newval
, curval
;
1340 int ret
, lock_taken
, ownerdied
= 0, attempt
= 0;
1342 if (refill_pi_state_cache())
1347 hrtimer_init_on_stack(&to
->timer
, CLOCK_REALTIME
,
1349 hrtimer_init_sleeper(to
, current
);
1350 hrtimer_set_expires(&to
->timer
, *time
);
1355 q
.key
= FUTEX_KEY_INIT
;
1356 ret
= get_futex_key(uaddr
, fshared
, &q
.key
);
1357 if (unlikely(ret
!= 0))
1358 goto out_release_sem
;
1361 hb
= queue_lock(&q
);
1364 ret
= lock_taken
= 0;
1367 * To avoid races, we attempt to take the lock here again
1368 * (by doing a 0 -> TID atomic cmpxchg), while holding all
1369 * the locks. It will most likely not succeed.
1371 newval
= task_pid_vnr(current
);
1373 curval
= cmpxchg_futex_value_locked(uaddr
, 0, newval
);
1375 if (unlikely(curval
== -EFAULT
))
1379 * Detect deadlocks. In case of REQUEUE_PI this is a valid
1380 * situation and we return success to user space.
1382 if (unlikely((curval
& FUTEX_TID_MASK
) == task_pid_vnr(current
))) {
1384 goto out_unlock_release_sem
;
1388 * Surprise - we got the lock. Just return to userspace:
1390 if (unlikely(!curval
))
1391 goto out_unlock_release_sem
;
1396 * Set the WAITERS flag, so the owner will know it has someone
1397 * to wake at next unlock
1399 newval
= curval
| FUTEX_WAITERS
;
1402 * There are two cases, where a futex might have no owner (the
1403 * owner TID is 0): OWNER_DIED. We take over the futex in this
1404 * case. We also do an unconditional take over, when the owner
1405 * of the futex died.
1407 * This is safe as we are protected by the hash bucket lock !
1409 if (unlikely(ownerdied
|| !(curval
& FUTEX_TID_MASK
))) {
1410 /* Keep the OWNER_DIED bit */
1411 newval
= (curval
& ~FUTEX_TID_MASK
) | task_pid_vnr(current
);
1416 curval
= cmpxchg_futex_value_locked(uaddr
, uval
, newval
);
1418 if (unlikely(curval
== -EFAULT
))
1420 if (unlikely(curval
!= uval
))
1424 * We took the lock due to owner died take over.
1426 if (unlikely(lock_taken
))
1427 goto out_unlock_release_sem
;
1430 * We dont have the lock. Look up the PI state (or create it if
1431 * we are the first waiter):
1433 ret
= lookup_pi_state(uval
, hb
, &q
.key
, &q
.pi_state
);
1435 if (unlikely(ret
)) {
1440 * Task is exiting and we just wait for the
1443 queue_unlock(&q
, hb
);
1449 * No owner found for this futex. Check if the
1450 * OWNER_DIED bit is set to figure out whether
1451 * this is a robust futex or not.
1453 if (get_futex_value_locked(&curval
, uaddr
))
1457 * We simply start over in case of a robust
1458 * futex. The code above will take the futex
1461 if (curval
& FUTEX_OWNER_DIED
) {
1466 goto out_unlock_release_sem
;
1471 * Only actually queue now that the atomic ops are done:
1475 WARN_ON(!q
.pi_state
);
1477 * Block on the PI mutex:
1480 ret
= rt_mutex_timed_lock(&q
.pi_state
->pi_mutex
, to
, 1);
1482 ret
= rt_mutex_trylock(&q
.pi_state
->pi_mutex
);
1483 /* Fixup the trylock return value: */
1484 ret
= ret
? 0 : -EWOULDBLOCK
;
1487 spin_lock(q
.lock_ptr
);
1491 * Got the lock. We might not be the anticipated owner
1492 * if we did a lock-steal - fix up the PI-state in
1495 if (q
.pi_state
->owner
!= curr
)
1496 ret
= fixup_pi_state_owner(uaddr
, &q
, curr
, fshared
);
1499 * Catch the rare case, where the lock was released
1500 * when we were on the way back before we locked the
1503 if (q
.pi_state
->owner
== curr
) {
1505 * Try to get the rt_mutex now. This might
1506 * fail as some other task acquired the
1507 * rt_mutex after we removed ourself from the
1508 * rt_mutex waiters list.
1510 if (rt_mutex_trylock(&q
.pi_state
->pi_mutex
))
1514 * pi_state is incorrect, some other
1515 * task did a lock steal and we
1516 * returned due to timeout or signal
1517 * without taking the rt_mutex. Too
1518 * late. We can access the
1519 * rt_mutex_owner without locking, as
1520 * the other task is now blocked on
1521 * the hash bucket lock. Fix the state
1524 struct task_struct
*owner
;
1527 owner
= rt_mutex_owner(&q
.pi_state
->pi_mutex
);
1528 res
= fixup_pi_state_owner(uaddr
, &q
, owner
,
1531 /* propagate -EFAULT, if the fixup failed */
1537 * Paranoia check. If we did not take the lock
1538 * in the trylock above, then we should not be
1539 * the owner of the rtmutex, neither the real
1540 * nor the pending one:
1542 if (rt_mutex_owner(&q
.pi_state
->pi_mutex
) == curr
)
1543 printk(KERN_ERR
"futex_lock_pi: ret = %d "
1544 "pi-mutex: %p pi-state %p\n", ret
,
1545 q
.pi_state
->pi_mutex
.owner
,
1550 /* Unqueue and drop the lock */
1554 destroy_hrtimer_on_stack(&to
->timer
);
1555 return ret
!= -EINTR
? ret
: -ERESTARTNOINTR
;
1557 out_unlock_release_sem
:
1558 queue_unlock(&q
, hb
);
1561 put_futex_key(fshared
, &q
.key
);
1563 destroy_hrtimer_on_stack(&to
->timer
);
1568 * We have to r/w *(int __user *)uaddr, and we have to modify it
1569 * atomically. Therefore, if we continue to fault after get_user()
1570 * below, we need to handle the fault ourselves, while still holding
1571 * the mmap_sem. This can occur if the uaddr is under contention as
1572 * we have to drop the mmap_sem in order to call get_user().
1574 queue_unlock(&q
, hb
);
1577 ret
= futex_handle_fault((unsigned long)uaddr
, attempt
);
1579 goto out_release_sem
;
1580 goto retry_unlocked
;
1583 ret
= get_user(uval
, uaddr
);
1588 destroy_hrtimer_on_stack(&to
->timer
);
1593 * Userspace attempted a TID -> 0 atomic transition, and failed.
1594 * This is the in-kernel slowpath: we look up the PI state (if any),
1595 * and do the rt-mutex unlock.
1597 static int futex_unlock_pi(u32 __user
*uaddr
, int fshared
)
1599 struct futex_hash_bucket
*hb
;
1600 struct futex_q
*this, *next
;
1602 struct plist_head
*head
;
1603 union futex_key key
= FUTEX_KEY_INIT
;
1604 int ret
, attempt
= 0;
1607 if (get_user(uval
, uaddr
))
1610 * We release only a lock we actually own:
1612 if ((uval
& FUTEX_TID_MASK
) != task_pid_vnr(current
))
1615 ret
= get_futex_key(uaddr
, fshared
, &key
);
1616 if (unlikely(ret
!= 0))
1619 hb
= hash_futex(&key
);
1621 spin_lock(&hb
->lock
);
1624 * To avoid races, try to do the TID -> 0 atomic transition
1625 * again. If it succeeds then we can return without waking
1628 if (!(uval
& FUTEX_OWNER_DIED
))
1629 uval
= cmpxchg_futex_value_locked(uaddr
, task_pid_vnr(current
), 0);
1632 if (unlikely(uval
== -EFAULT
))
1635 * Rare case: we managed to release the lock atomically,
1636 * no need to wake anyone else up:
1638 if (unlikely(uval
== task_pid_vnr(current
)))
1642 * Ok, other tasks may need to be woken up - check waiters
1643 * and do the wakeup if necessary:
1647 plist_for_each_entry_safe(this, next
, head
, list
) {
1648 if (!match_futex (&this->key
, &key
))
1650 ret
= wake_futex_pi(uaddr
, uval
, this);
1652 * The atomic access to the futex value
1653 * generated a pagefault, so retry the
1654 * user-access and the wakeup:
1661 * No waiters - kernel unlocks the futex:
1663 if (!(uval
& FUTEX_OWNER_DIED
)) {
1664 ret
= unlock_futex_pi(uaddr
, uval
);
1670 spin_unlock(&hb
->lock
);
1672 put_futex_key(fshared
, &key
);
1678 * We have to r/w *(int __user *)uaddr, and we have to modify it
1679 * atomically. Therefore, if we continue to fault after get_user()
1680 * below, we need to handle the fault ourselves, while still holding
1681 * the mmap_sem. This can occur if the uaddr is under contention as
1682 * we have to drop the mmap_sem in order to call get_user().
1684 spin_unlock(&hb
->lock
);
1687 ret
= futex_handle_fault((unsigned long)uaddr
, attempt
);
1691 goto retry_unlocked
;
1694 ret
= get_user(uval
, uaddr
);
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
, 1, 1, FUTEX_BITSET_MATCH_ANY
);
1825 * Fetch a robust-list pointer. Bit 0 signals PI futexes:
1827 static inline int fetch_robust_entry(struct robust_list __user
**entry
,
1828 struct robust_list __user
* __user
*head
,
1831 unsigned long uentry
;
1833 if (get_user(uentry
, (unsigned long __user
*)head
))
1836 *entry
= (void __user
*)(uentry
& ~1UL);
1843 * Walk curr->robust_list (very carefully, it's a userspace list!)
1844 * and mark any locks found there dead, and notify any waiters.
1846 * We silently return on any sign of list-walking problem.
1848 void exit_robust_list(struct task_struct
*curr
)
1850 struct robust_list_head __user
*head
= curr
->robust_list
;
1851 struct robust_list __user
*entry
, *next_entry
, *pending
;
1852 unsigned int limit
= ROBUST_LIST_LIMIT
, pi
, next_pi
, pip
;
1853 unsigned long futex_offset
;
1856 if (!futex_cmpxchg_enabled
)
1860 * Fetch the list head (which was registered earlier, via
1861 * sys_set_robust_list()):
1863 if (fetch_robust_entry(&entry
, &head
->list
.next
, &pi
))
1866 * Fetch the relative futex offset:
1868 if (get_user(futex_offset
, &head
->futex_offset
))
1871 * Fetch any possibly pending lock-add first, and handle it
1874 if (fetch_robust_entry(&pending
, &head
->list_op_pending
, &pip
))
1877 next_entry
= NULL
; /* avoid warning with gcc */
1878 while (entry
!= &head
->list
) {
1880 * Fetch the next entry in the list before calling
1881 * handle_futex_death:
1883 rc
= fetch_robust_entry(&next_entry
, &entry
->next
, &next_pi
);
1885 * A pending lock might already be on the list, so
1886 * don't process it twice:
1888 if (entry
!= pending
)
1889 if (handle_futex_death((void __user
*)entry
+ futex_offset
,
1897 * Avoid excessively long or circular lists:
1906 handle_futex_death((void __user
*)pending
+ futex_offset
,
1910 long do_futex(u32 __user
*uaddr
, int op
, u32 val
, ktime_t
*timeout
,
1911 u32 __user
*uaddr2
, u32 val2
, u32 val3
)
1913 int clockrt
, ret
= -ENOSYS
;
1914 int cmd
= op
& FUTEX_CMD_MASK
;
1917 if (!(op
& FUTEX_PRIVATE_FLAG
))
1920 clockrt
= op
& FUTEX_CLOCK_REALTIME
;
1921 if (clockrt
&& cmd
!= FUTEX_WAIT_BITSET
)
1926 val3
= FUTEX_BITSET_MATCH_ANY
;
1927 case FUTEX_WAIT_BITSET
:
1928 ret
= futex_wait(uaddr
, fshared
, val
, timeout
, val3
, clockrt
);
1931 val3
= FUTEX_BITSET_MATCH_ANY
;
1932 case FUTEX_WAKE_BITSET
:
1933 ret
= futex_wake(uaddr
, fshared
, val
, val3
);
1936 ret
= futex_requeue(uaddr
, fshared
, uaddr2
, val
, val2
, NULL
);
1938 case FUTEX_CMP_REQUEUE
:
1939 ret
= futex_requeue(uaddr
, fshared
, uaddr2
, val
, val2
, &val3
);
1942 ret
= futex_wake_op(uaddr
, fshared
, uaddr2
, val
, val2
, val3
);
1945 if (futex_cmpxchg_enabled
)
1946 ret
= futex_lock_pi(uaddr
, fshared
, val
, timeout
, 0);
1948 case FUTEX_UNLOCK_PI
:
1949 if (futex_cmpxchg_enabled
)
1950 ret
= futex_unlock_pi(uaddr
, fshared
);
1952 case FUTEX_TRYLOCK_PI
:
1953 if (futex_cmpxchg_enabled
)
1954 ret
= futex_lock_pi(uaddr
, fshared
, 0, timeout
, 1);
1963 asmlinkage
long sys_futex(u32 __user
*uaddr
, int op
, u32 val
,
1964 struct timespec __user
*utime
, u32 __user
*uaddr2
,
1968 ktime_t t
, *tp
= NULL
;
1970 int cmd
= op
& FUTEX_CMD_MASK
;
1972 if (utime
&& (cmd
== FUTEX_WAIT
|| cmd
== FUTEX_LOCK_PI
||
1973 cmd
== FUTEX_WAIT_BITSET
)) {
1974 if (copy_from_user(&ts
, utime
, sizeof(ts
)) != 0)
1976 if (!timespec_valid(&ts
))
1979 t
= timespec_to_ktime(ts
);
1980 if (cmd
== FUTEX_WAIT
)
1981 t
= ktime_add_safe(ktime_get(), t
);
1985 * requeue parameter in 'utime' if cmd == FUTEX_REQUEUE.
1986 * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP.
1988 if (cmd
== FUTEX_REQUEUE
|| cmd
== FUTEX_CMP_REQUEUE
||
1989 cmd
== FUTEX_WAKE_OP
)
1990 val2
= (u32
) (unsigned long) utime
;
1992 return do_futex(uaddr
, op
, val
, tp
, uaddr2
, val2
, val3
);
1995 static int __init
futex_init(void)
2001 * This will fail and we want it. Some arch implementations do
2002 * runtime detection of the futex_atomic_cmpxchg_inatomic()
2003 * functionality. We want to know that before we call in any
2004 * of the complex code paths. Also we want to prevent
2005 * registration of robust lists in that case. NULL is
2006 * guaranteed to fault and we get -EFAULT on functional
2007 * implementation, the non functional ones will return
2010 curval
= cmpxchg_futex_value_locked(NULL
, 0, 0);
2011 if (curval
== -EFAULT
)
2012 futex_cmpxchg_enabled
= 1;
2014 for (i
= 0; i
< ARRAY_SIZE(futex_queues
); i
++) {
2015 plist_head_init(&futex_queues
[i
].chain
, &futex_queues
[i
].lock
);
2016 spin_lock_init(&futex_queues
[i
].lock
);
2021 __initcall(futex_init
);