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
, int fshared
, union futex_key
*key
)
205 unsigned long address
= (unsigned long)uaddr
;
206 struct mm_struct
*mm
= current
->mm
;
211 * The futex address must be "naturally" aligned.
213 key
->both
.offset
= address
% PAGE_SIZE
;
214 if (unlikely((address
% sizeof(u32
)) != 0))
216 address
-= key
->both
.offset
;
219 * PROCESS_PRIVATE futexes are fast.
220 * As the mm cannot disappear under us and the 'key' only needs
221 * virtual address, we dont even have to find the underlying vma.
222 * Note : We do have to check 'uaddr' is a valid user address,
223 * but access_ok() should be faster than find_vma()
226 if (unlikely(!access_ok(VERIFY_WRITE
, uaddr
, sizeof(u32
))))
228 key
->private.mm
= mm
;
229 key
->private.address
= address
;
230 get_futex_key_refs(key
);
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(int 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;
301 static int futex_handle_fault(unsigned long address
, int attempt
)
303 struct vm_area_struct
* vma
;
304 struct mm_struct
*mm
= current
->mm
;
310 down_read(&mm
->mmap_sem
);
311 vma
= find_vma(mm
, address
);
312 if (vma
&& address
>= vma
->vm_start
&&
313 (vma
->vm_flags
& VM_WRITE
)) {
315 fault
= handle_mm_fault(mm
, vma
, address
, 1);
316 if (unlikely((fault
& VM_FAULT_ERROR
))) {
318 /* XXX: let's do this when we verify it is OK */
319 if (ret
& VM_FAULT_OOM
)
324 if (fault
& VM_FAULT_MAJOR
)
330 up_read(&mm
->mmap_sem
);
337 static int refill_pi_state_cache(void)
339 struct futex_pi_state
*pi_state
;
341 if (likely(current
->pi_state_cache
))
344 pi_state
= kzalloc(sizeof(*pi_state
), GFP_KERNEL
);
349 INIT_LIST_HEAD(&pi_state
->list
);
350 /* pi_mutex gets initialized later */
351 pi_state
->owner
= NULL
;
352 atomic_set(&pi_state
->refcount
, 1);
353 pi_state
->key
= FUTEX_KEY_INIT
;
355 current
->pi_state_cache
= pi_state
;
360 static struct futex_pi_state
* alloc_pi_state(void)
362 struct futex_pi_state
*pi_state
= current
->pi_state_cache
;
365 current
->pi_state_cache
= NULL
;
370 static void free_pi_state(struct futex_pi_state
*pi_state
)
372 if (!atomic_dec_and_test(&pi_state
->refcount
))
376 * If pi_state->owner is NULL, the owner is most probably dying
377 * and has cleaned up the pi_state already
379 if (pi_state
->owner
) {
380 spin_lock_irq(&pi_state
->owner
->pi_lock
);
381 list_del_init(&pi_state
->list
);
382 spin_unlock_irq(&pi_state
->owner
->pi_lock
);
384 rt_mutex_proxy_unlock(&pi_state
->pi_mutex
, pi_state
->owner
);
387 if (current
->pi_state_cache
)
391 * pi_state->list is already empty.
392 * clear pi_state->owner.
393 * refcount is at 0 - put it back to 1.
395 pi_state
->owner
= NULL
;
396 atomic_set(&pi_state
->refcount
, 1);
397 current
->pi_state_cache
= pi_state
;
402 * Look up the task based on what TID userspace gave us.
405 static struct task_struct
* futex_find_get_task(pid_t pid
)
407 struct task_struct
*p
;
410 p
= find_task_by_vpid(pid
);
411 if (!p
|| ((current
->euid
!= p
->euid
) && (current
->euid
!= p
->uid
)))
422 * This task is holding PI mutexes at exit time => bad.
423 * Kernel cleans up PI-state, but userspace is likely hosed.
424 * (Robust-futex cleanup is separate and might save the day for userspace.)
426 void exit_pi_state_list(struct task_struct
*curr
)
428 struct list_head
*next
, *head
= &curr
->pi_state_list
;
429 struct futex_pi_state
*pi_state
;
430 struct futex_hash_bucket
*hb
;
431 union futex_key key
= FUTEX_KEY_INIT
;
433 if (!futex_cmpxchg_enabled
)
436 * We are a ZOMBIE and nobody can enqueue itself on
437 * pi_state_list anymore, but we have to be careful
438 * versus waiters unqueueing themselves:
440 spin_lock_irq(&curr
->pi_lock
);
441 while (!list_empty(head
)) {
444 pi_state
= list_entry(next
, struct futex_pi_state
, list
);
446 hb
= hash_futex(&key
);
447 spin_unlock_irq(&curr
->pi_lock
);
449 spin_lock(&hb
->lock
);
451 spin_lock_irq(&curr
->pi_lock
);
453 * We dropped the pi-lock, so re-check whether this
454 * task still owns the PI-state:
456 if (head
->next
!= next
) {
457 spin_unlock(&hb
->lock
);
461 WARN_ON(pi_state
->owner
!= curr
);
462 WARN_ON(list_empty(&pi_state
->list
));
463 list_del_init(&pi_state
->list
);
464 pi_state
->owner
= NULL
;
465 spin_unlock_irq(&curr
->pi_lock
);
467 rt_mutex_unlock(&pi_state
->pi_mutex
);
469 spin_unlock(&hb
->lock
);
471 spin_lock_irq(&curr
->pi_lock
);
473 spin_unlock_irq(&curr
->pi_lock
);
477 lookup_pi_state(u32 uval
, struct futex_hash_bucket
*hb
,
478 union futex_key
*key
, struct futex_pi_state
**ps
)
480 struct futex_pi_state
*pi_state
= NULL
;
481 struct futex_q
*this, *next
;
482 struct plist_head
*head
;
483 struct task_struct
*p
;
484 pid_t pid
= uval
& FUTEX_TID_MASK
;
488 plist_for_each_entry_safe(this, next
, head
, list
) {
489 if (match_futex(&this->key
, key
)) {
491 * Another waiter already exists - bump up
492 * the refcount and return its pi_state:
494 pi_state
= this->pi_state
;
496 * Userspace might have messed up non PI and PI futexes
498 if (unlikely(!pi_state
))
501 WARN_ON(!atomic_read(&pi_state
->refcount
));
502 WARN_ON(pid
&& pi_state
->owner
&&
503 pi_state
->owner
->pid
!= pid
);
505 atomic_inc(&pi_state
->refcount
);
513 * We are the first waiter - try to look up the real owner and attach
514 * the new pi_state to it, but bail out when TID = 0
518 p
= futex_find_get_task(pid
);
523 * We need to look at the task state flags to figure out,
524 * whether the task is exiting. To protect against the do_exit
525 * change of the task flags, we do this protected by
528 spin_lock_irq(&p
->pi_lock
);
529 if (unlikely(p
->flags
& PF_EXITING
)) {
531 * The task is on the way out. When PF_EXITPIDONE is
532 * set, we know that the task has finished the
535 int ret
= (p
->flags
& PF_EXITPIDONE
) ? -ESRCH
: -EAGAIN
;
537 spin_unlock_irq(&p
->pi_lock
);
542 pi_state
= alloc_pi_state();
545 * Initialize the pi_mutex in locked state and make 'p'
548 rt_mutex_init_proxy_locked(&pi_state
->pi_mutex
, p
);
550 /* Store the key for possible exit cleanups: */
551 pi_state
->key
= *key
;
553 WARN_ON(!list_empty(&pi_state
->list
));
554 list_add(&pi_state
->list
, &p
->pi_state_list
);
556 spin_unlock_irq(&p
->pi_lock
);
566 * The hash bucket lock must be held when this is called.
567 * Afterwards, the futex_q must not be accessed.
569 static void wake_futex(struct futex_q
*q
)
571 plist_del(&q
->list
, &q
->list
.plist
);
573 * The lock in wake_up_all() is a crucial memory barrier after the
574 * plist_del() and also before assigning to q->lock_ptr.
576 wake_up_all(&q
->waiters
);
578 * The waiting task can free the futex_q as soon as this is written,
579 * without taking any locks. This must come last.
581 * A memory barrier is required here to prevent the following store
582 * to lock_ptr from getting ahead of the wakeup. Clearing the lock
583 * at the end of wake_up_all() does not prevent this store from
590 static int wake_futex_pi(u32 __user
*uaddr
, u32 uval
, struct futex_q
*this)
592 struct task_struct
*new_owner
;
593 struct futex_pi_state
*pi_state
= this->pi_state
;
599 spin_lock(&pi_state
->pi_mutex
.wait_lock
);
600 new_owner
= rt_mutex_next_owner(&pi_state
->pi_mutex
);
603 * This happens when we have stolen the lock and the original
604 * pending owner did not enqueue itself back on the rt_mutex.
605 * Thats not a tragedy. We know that way, that a lock waiter
606 * is on the fly. We make the futex_q waiter the pending owner.
609 new_owner
= this->task
;
612 * We pass it to the next owner. (The WAITERS bit is always
613 * kept enabled while there is PI state around. We must also
614 * preserve the owner died bit.)
616 if (!(uval
& FUTEX_OWNER_DIED
)) {
619 newval
= FUTEX_WAITERS
| task_pid_vnr(new_owner
);
621 curval
= cmpxchg_futex_value_locked(uaddr
, uval
, newval
);
623 if (curval
== -EFAULT
)
625 else if (curval
!= uval
)
628 spin_unlock(&pi_state
->pi_mutex
.wait_lock
);
633 spin_lock_irq(&pi_state
->owner
->pi_lock
);
634 WARN_ON(list_empty(&pi_state
->list
));
635 list_del_init(&pi_state
->list
);
636 spin_unlock_irq(&pi_state
->owner
->pi_lock
);
638 spin_lock_irq(&new_owner
->pi_lock
);
639 WARN_ON(!list_empty(&pi_state
->list
));
640 list_add(&pi_state
->list
, &new_owner
->pi_state_list
);
641 pi_state
->owner
= new_owner
;
642 spin_unlock_irq(&new_owner
->pi_lock
);
644 spin_unlock(&pi_state
->pi_mutex
.wait_lock
);
645 rt_mutex_unlock(&pi_state
->pi_mutex
);
650 static int unlock_futex_pi(u32 __user
*uaddr
, u32 uval
)
655 * There is no waiter, so we unlock the futex. The owner died
656 * bit has not to be preserved here. We are the owner:
658 oldval
= cmpxchg_futex_value_locked(uaddr
, uval
, 0);
660 if (oldval
== -EFAULT
)
669 * Express the locking dependencies for lockdep:
672 double_lock_hb(struct futex_hash_bucket
*hb1
, struct futex_hash_bucket
*hb2
)
675 spin_lock(&hb1
->lock
);
677 spin_lock_nested(&hb2
->lock
, SINGLE_DEPTH_NESTING
);
678 } else { /* hb1 > hb2 */
679 spin_lock(&hb2
->lock
);
680 spin_lock_nested(&hb1
->lock
, SINGLE_DEPTH_NESTING
);
685 * Wake up all waiters hashed on the physical page that is mapped
686 * to this virtual address:
688 static int futex_wake(u32 __user
*uaddr
, int fshared
, int nr_wake
, u32 bitset
)
690 struct futex_hash_bucket
*hb
;
691 struct futex_q
*this, *next
;
692 struct plist_head
*head
;
693 union futex_key key
= FUTEX_KEY_INIT
;
699 ret
= get_futex_key(uaddr
, fshared
, &key
);
700 if (unlikely(ret
!= 0))
703 hb
= hash_futex(&key
);
704 spin_lock(&hb
->lock
);
707 plist_for_each_entry_safe(this, next
, head
, list
) {
708 if (match_futex (&this->key
, &key
)) {
709 if (this->pi_state
) {
714 /* Check if one of the bits is set in both bitsets */
715 if (!(this->bitset
& bitset
))
719 if (++ret
>= nr_wake
)
724 spin_unlock(&hb
->lock
);
726 put_futex_key(fshared
, &key
);
731 * Wake up all waiters hashed on the physical page that is mapped
732 * to this virtual address:
735 futex_wake_op(u32 __user
*uaddr1
, int fshared
, u32 __user
*uaddr2
,
736 int nr_wake
, int nr_wake2
, int op
)
738 union futex_key key1
= FUTEX_KEY_INIT
, key2
= FUTEX_KEY_INIT
;
739 struct futex_hash_bucket
*hb1
, *hb2
;
740 struct plist_head
*head
;
741 struct futex_q
*this, *next
;
742 int ret
, op_ret
, attempt
= 0;
745 ret
= get_futex_key(uaddr1
, fshared
, &key1
);
746 if (unlikely(ret
!= 0))
748 ret
= get_futex_key(uaddr2
, fshared
, &key2
);
749 if (unlikely(ret
!= 0))
752 hb1
= hash_futex(&key1
);
753 hb2
= hash_futex(&key2
);
756 double_lock_hb(hb1
, hb2
);
758 op_ret
= futex_atomic_op_inuser(op
, uaddr2
);
759 if (unlikely(op_ret
< 0)) {
762 spin_unlock(&hb1
->lock
);
764 spin_unlock(&hb2
->lock
);
768 * we don't get EFAULT from MMU faults if we don't have an MMU,
769 * but we might get them from range checking
775 if (unlikely(op_ret
!= -EFAULT
)) {
781 * futex_atomic_op_inuser needs to both read and write
782 * *(int __user *)uaddr2, but we can't modify it
783 * non-atomically. Therefore, if get_user below is not
784 * enough, we need to handle the fault ourselves, while
785 * still holding the mmap_sem.
788 ret
= futex_handle_fault((unsigned long)uaddr2
,
795 ret
= get_user(dummy
, uaddr2
);
804 plist_for_each_entry_safe(this, next
, head
, list
) {
805 if (match_futex (&this->key
, &key1
)) {
807 if (++ret
>= nr_wake
)
816 plist_for_each_entry_safe(this, next
, head
, list
) {
817 if (match_futex (&this->key
, &key2
)) {
819 if (++op_ret
>= nr_wake2
)
826 spin_unlock(&hb1
->lock
);
828 spin_unlock(&hb2
->lock
);
830 put_futex_key(fshared
, &key2
);
831 put_futex_key(fshared
, &key1
);
837 * Requeue all waiters hashed on one physical page to another
840 static int futex_requeue(u32 __user
*uaddr1
, int fshared
, u32 __user
*uaddr2
,
841 int nr_wake
, int nr_requeue
, u32
*cmpval
)
843 union futex_key key1
= FUTEX_KEY_INIT
, key2
= FUTEX_KEY_INIT
;
844 struct futex_hash_bucket
*hb1
, *hb2
;
845 struct plist_head
*head1
;
846 struct futex_q
*this, *next
;
847 int ret
, drop_count
= 0;
850 ret
= get_futex_key(uaddr1
, fshared
, &key1
);
851 if (unlikely(ret
!= 0))
853 ret
= get_futex_key(uaddr2
, fshared
, &key2
);
854 if (unlikely(ret
!= 0))
857 hb1
= hash_futex(&key1
);
858 hb2
= hash_futex(&key2
);
860 double_lock_hb(hb1
, hb2
);
862 if (likely(cmpval
!= NULL
)) {
865 ret
= get_futex_value_locked(&curval
, uaddr1
);
868 spin_unlock(&hb1
->lock
);
870 spin_unlock(&hb2
->lock
);
872 ret
= get_user(curval
, uaddr1
);
879 if (curval
!= *cmpval
) {
886 plist_for_each_entry_safe(this, next
, head1
, list
) {
887 if (!match_futex (&this->key
, &key1
))
889 if (++ret
<= nr_wake
) {
893 * If key1 and key2 hash to the same bucket, no need to
896 if (likely(head1
!= &hb2
->chain
)) {
897 plist_del(&this->list
, &hb1
->chain
);
898 plist_add(&this->list
, &hb2
->chain
);
899 this->lock_ptr
= &hb2
->lock
;
900 #ifdef CONFIG_DEBUG_PI_LIST
901 this->list
.plist
.lock
= &hb2
->lock
;
905 get_futex_key_refs(&key2
);
908 if (ret
- nr_wake
>= nr_requeue
)
914 spin_unlock(&hb1
->lock
);
916 spin_unlock(&hb2
->lock
);
918 /* drop_futex_key_refs() must be called outside the spinlocks. */
919 while (--drop_count
>= 0)
920 drop_futex_key_refs(&key1
);
923 put_futex_key(fshared
, &key2
);
924 put_futex_key(fshared
, &key1
);
928 /* The key must be already stored in q->key. */
929 static inline struct futex_hash_bucket
*queue_lock(struct futex_q
*q
)
931 struct futex_hash_bucket
*hb
;
933 init_waitqueue_head(&q
->waiters
);
935 get_futex_key_refs(&q
->key
);
936 hb
= hash_futex(&q
->key
);
937 q
->lock_ptr
= &hb
->lock
;
939 spin_lock(&hb
->lock
);
943 static inline void queue_me(struct futex_q
*q
, struct futex_hash_bucket
*hb
)
948 * The priority used to register this element is
949 * - either the real thread-priority for the real-time threads
950 * (i.e. threads with a priority lower than MAX_RT_PRIO)
951 * - or MAX_RT_PRIO for non-RT threads.
952 * Thus, all RT-threads are woken first in priority order, and
953 * the others are woken last, in FIFO order.
955 prio
= min(current
->normal_prio
, MAX_RT_PRIO
);
957 plist_node_init(&q
->list
, prio
);
958 #ifdef CONFIG_DEBUG_PI_LIST
959 q
->list
.plist
.lock
= &hb
->lock
;
961 plist_add(&q
->list
, &hb
->chain
);
963 spin_unlock(&hb
->lock
);
967 queue_unlock(struct futex_q
*q
, struct futex_hash_bucket
*hb
)
969 spin_unlock(&hb
->lock
);
970 drop_futex_key_refs(&q
->key
);
974 * queue_me and unqueue_me must be called as a pair, each
975 * exactly once. They are called with the hashed spinlock held.
978 /* Return 1 if we were still queued (ie. 0 means we were woken) */
979 static int unqueue_me(struct futex_q
*q
)
981 spinlock_t
*lock_ptr
;
984 /* In the common case we don't take the spinlock, which is nice. */
986 lock_ptr
= q
->lock_ptr
;
988 if (lock_ptr
!= NULL
) {
991 * q->lock_ptr can change between reading it and
992 * spin_lock(), causing us to take the wrong lock. This
993 * corrects the race condition.
995 * Reasoning goes like this: if we have the wrong lock,
996 * q->lock_ptr must have changed (maybe several times)
997 * between reading it and the spin_lock(). It can
998 * change again after the spin_lock() but only if it was
999 * already changed before the spin_lock(). It cannot,
1000 * however, change back to the original value. Therefore
1001 * we can detect whether we acquired the correct lock.
1003 if (unlikely(lock_ptr
!= q
->lock_ptr
)) {
1004 spin_unlock(lock_ptr
);
1007 WARN_ON(plist_node_empty(&q
->list
));
1008 plist_del(&q
->list
, &q
->list
.plist
);
1010 BUG_ON(q
->pi_state
);
1012 spin_unlock(lock_ptr
);
1016 drop_futex_key_refs(&q
->key
);
1021 * PI futexes can not be requeued and must remove themself from the
1022 * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry
1025 static void unqueue_me_pi(struct futex_q
*q
)
1027 WARN_ON(plist_node_empty(&q
->list
));
1028 plist_del(&q
->list
, &q
->list
.plist
);
1030 BUG_ON(!q
->pi_state
);
1031 free_pi_state(q
->pi_state
);
1034 spin_unlock(q
->lock_ptr
);
1036 drop_futex_key_refs(&q
->key
);
1040 * Fixup the pi_state owner with the new owner.
1042 * Must be called with hash bucket lock held and mm->sem held for non
1045 static int fixup_pi_state_owner(u32 __user
*uaddr
, struct futex_q
*q
,
1046 struct task_struct
*newowner
, int fshared
)
1048 u32 newtid
= task_pid_vnr(newowner
) | FUTEX_WAITERS
;
1049 struct futex_pi_state
*pi_state
= q
->pi_state
;
1050 struct task_struct
*oldowner
= pi_state
->owner
;
1051 u32 uval
, curval
, newval
;
1052 int ret
, attempt
= 0;
1055 if (!pi_state
->owner
)
1056 newtid
|= FUTEX_OWNER_DIED
;
1059 * We are here either because we stole the rtmutex from the
1060 * pending owner or we are the pending owner which failed to
1061 * get the rtmutex. We have to replace the pending owner TID
1062 * in the user space variable. This must be atomic as we have
1063 * to preserve the owner died bit here.
1065 * Note: We write the user space value _before_ changing the
1066 * pi_state because we can fault here. Imagine swapped out
1067 * pages or a fork, which was running right before we acquired
1068 * mmap_sem, that marked all the anonymous memory readonly for
1071 * Modifying pi_state _before_ the user space value would
1072 * leave the pi_state in an inconsistent state when we fault
1073 * here, because we need to drop the hash bucket lock to
1074 * handle the fault. This might be observed in the PID check
1075 * in lookup_pi_state.
1078 if (get_futex_value_locked(&uval
, uaddr
))
1082 newval
= (uval
& FUTEX_OWNER_DIED
) | newtid
;
1084 curval
= cmpxchg_futex_value_locked(uaddr
, uval
, newval
);
1086 if (curval
== -EFAULT
)
1094 * We fixed up user space. Now we need to fix the pi_state
1097 if (pi_state
->owner
!= NULL
) {
1098 spin_lock_irq(&pi_state
->owner
->pi_lock
);
1099 WARN_ON(list_empty(&pi_state
->list
));
1100 list_del_init(&pi_state
->list
);
1101 spin_unlock_irq(&pi_state
->owner
->pi_lock
);
1104 pi_state
->owner
= newowner
;
1106 spin_lock_irq(&newowner
->pi_lock
);
1107 WARN_ON(!list_empty(&pi_state
->list
));
1108 list_add(&pi_state
->list
, &newowner
->pi_state_list
);
1109 spin_unlock_irq(&newowner
->pi_lock
);
1113 * To handle the page fault we need to drop the hash bucket
1114 * lock here. That gives the other task (either the pending
1115 * owner itself or the task which stole the rtmutex) the
1116 * chance to try the fixup of the pi_state. So once we are
1117 * back from handling the fault we need to check the pi_state
1118 * after reacquiring the hash bucket lock and before trying to
1119 * do another fixup. When the fixup has been done already we
1123 spin_unlock(q
->lock_ptr
);
1125 ret
= futex_handle_fault((unsigned long)uaddr
, attempt
++);
1127 spin_lock(q
->lock_ptr
);
1130 * Check if someone else fixed it for us:
1132 if (pi_state
->owner
!= oldowner
)
1142 * In case we must use restart_block to restart a futex_wait,
1143 * we encode in the 'flags' shared capability
1145 #define FLAGS_SHARED 1
1147 static long futex_wait_restart(struct restart_block
*restart
);
1149 static int futex_wait(u32 __user
*uaddr
, int fshared
,
1150 u32 val
, ktime_t
*abs_time
, u32 bitset
)
1152 struct task_struct
*curr
= current
;
1153 DECLARE_WAITQUEUE(wait
, curr
);
1154 struct futex_hash_bucket
*hb
;
1158 struct hrtimer_sleeper t
;
1167 q
.key
= FUTEX_KEY_INIT
;
1168 ret
= get_futex_key(uaddr
, fshared
, &q
.key
);
1169 if (unlikely(ret
!= 0))
1170 goto out_release_sem
;
1172 hb
= queue_lock(&q
);
1175 * Access the page AFTER the futex is queued.
1176 * Order is important:
1178 * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val);
1179 * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); }
1181 * The basic logical guarantee of a futex is that it blocks ONLY
1182 * if cond(var) is known to be true at the time of blocking, for
1183 * any cond. If we queued after testing *uaddr, that would open
1184 * a race condition where we could block indefinitely with
1185 * cond(var) false, which would violate the guarantee.
1187 * A consequence is that futex_wait() can return zero and absorb
1188 * a wakeup when *uaddr != val on entry to the syscall. This is
1191 * for shared futexes, we hold the mmap semaphore, so the mapping
1192 * cannot have changed since we looked it up in get_futex_key.
1194 ret
= get_futex_value_locked(&uval
, uaddr
);
1196 if (unlikely(ret
)) {
1197 queue_unlock(&q
, hb
);
1199 ret
= get_user(uval
, uaddr
);
1207 goto out_unlock_release_sem
;
1209 /* Only actually queue if *uaddr contained val. */
1213 * There might have been scheduling since the queue_me(), as we
1214 * cannot hold a spinlock across the get_user() in case it
1215 * faults, and we cannot just set TASK_INTERRUPTIBLE state when
1216 * queueing ourselves into the futex hash. This code thus has to
1217 * rely on the futex_wake() code removing us from hash when it
1221 /* add_wait_queue is the barrier after __set_current_state. */
1222 __set_current_state(TASK_INTERRUPTIBLE
);
1223 add_wait_queue(&q
.waiters
, &wait
);
1225 * !plist_node_empty() is safe here without any lock.
1226 * q.lock_ptr != 0 is not safe, because of ordering against wakeup.
1228 if (likely(!plist_node_empty(&q
.list
))) {
1232 unsigned long slack
;
1233 slack
= current
->timer_slack_ns
;
1234 if (rt_task(current
))
1236 hrtimer_init_on_stack(&t
.timer
, CLOCK_MONOTONIC
,
1238 hrtimer_init_sleeper(&t
, current
);
1239 hrtimer_set_expires_range_ns(&t
.timer
, *abs_time
, slack
);
1241 hrtimer_start_expires(&t
.timer
, HRTIMER_MODE_ABS
);
1242 if (!hrtimer_active(&t
.timer
))
1246 * the timer could have already expired, in which
1247 * case current would be flagged for rescheduling.
1248 * Don't bother calling schedule.
1253 hrtimer_cancel(&t
.timer
);
1255 /* Flag if a timeout occured */
1256 rem
= (t
.task
== NULL
);
1258 destroy_hrtimer_on_stack(&t
.timer
);
1261 __set_current_state(TASK_RUNNING
);
1264 * NOTE: we don't remove ourselves from the waitqueue because
1265 * we are the only user of it.
1268 /* If we were woken (and unqueued), we succeeded, whatever. */
1269 if (!unqueue_me(&q
))
1275 * We expect signal_pending(current), but another thread may
1276 * have handled it for us already.
1279 return -ERESTARTSYS
;
1281 struct restart_block
*restart
;
1282 restart
= ¤t_thread_info()->restart_block
;
1283 restart
->fn
= futex_wait_restart
;
1284 restart
->futex
.uaddr
= (u32
*)uaddr
;
1285 restart
->futex
.val
= val
;
1286 restart
->futex
.time
= abs_time
->tv64
;
1287 restart
->futex
.bitset
= bitset
;
1288 restart
->futex
.flags
= 0;
1291 restart
->futex
.flags
|= FLAGS_SHARED
;
1292 return -ERESTART_RESTARTBLOCK
;
1295 out_unlock_release_sem
:
1296 queue_unlock(&q
, hb
);
1299 put_futex_key(fshared
, &q
.key
);
1304 static long futex_wait_restart(struct restart_block
*restart
)
1306 u32 __user
*uaddr
= (u32 __user
*)restart
->futex
.uaddr
;
1310 t
.tv64
= restart
->futex
.time
;
1311 restart
->fn
= do_no_restart_syscall
;
1312 if (restart
->futex
.flags
& FLAGS_SHARED
)
1314 return (long)futex_wait(uaddr
, fshared
, restart
->futex
.val
, &t
,
1315 restart
->futex
.bitset
);
1320 * Userspace tried a 0 -> TID atomic transition of the futex value
1321 * and failed. The kernel side here does the whole locking operation:
1322 * if there are waiters then it will block, it does PI, etc. (Due to
1323 * races the kernel might see a 0 value of the futex too.)
1325 static int futex_lock_pi(u32 __user
*uaddr
, int fshared
,
1326 int detect
, ktime_t
*time
, int trylock
)
1328 struct hrtimer_sleeper timeout
, *to
= NULL
;
1329 struct task_struct
*curr
= current
;
1330 struct futex_hash_bucket
*hb
;
1331 u32 uval
, newval
, curval
;
1333 int ret
, lock_taken
, ownerdied
= 0, attempt
= 0;
1335 if (refill_pi_state_cache())
1340 hrtimer_init_on_stack(&to
->timer
, CLOCK_REALTIME
,
1342 hrtimer_init_sleeper(to
, current
);
1343 hrtimer_set_expires(&to
->timer
, *time
);
1348 q
.key
= FUTEX_KEY_INIT
;
1349 ret
= get_futex_key(uaddr
, fshared
, &q
.key
);
1350 if (unlikely(ret
!= 0))
1351 goto out_release_sem
;
1354 hb
= queue_lock(&q
);
1357 ret
= lock_taken
= 0;
1360 * To avoid races, we attempt to take the lock here again
1361 * (by doing a 0 -> TID atomic cmpxchg), while holding all
1362 * the locks. It will most likely not succeed.
1364 newval
= task_pid_vnr(current
);
1366 curval
= cmpxchg_futex_value_locked(uaddr
, 0, newval
);
1368 if (unlikely(curval
== -EFAULT
))
1372 * Detect deadlocks. In case of REQUEUE_PI this is a valid
1373 * situation and we return success to user space.
1375 if (unlikely((curval
& FUTEX_TID_MASK
) == task_pid_vnr(current
))) {
1377 goto out_unlock_release_sem
;
1381 * Surprise - we got the lock. Just return to userspace:
1383 if (unlikely(!curval
))
1384 goto out_unlock_release_sem
;
1389 * Set the WAITERS flag, so the owner will know it has someone
1390 * to wake at next unlock
1392 newval
= curval
| FUTEX_WAITERS
;
1395 * There are two cases, where a futex might have no owner (the
1396 * owner TID is 0): OWNER_DIED. We take over the futex in this
1397 * case. We also do an unconditional take over, when the owner
1398 * of the futex died.
1400 * This is safe as we are protected by the hash bucket lock !
1402 if (unlikely(ownerdied
|| !(curval
& FUTEX_TID_MASK
))) {
1403 /* Keep the OWNER_DIED bit */
1404 newval
= (curval
& ~FUTEX_TID_MASK
) | task_pid_vnr(current
);
1409 curval
= cmpxchg_futex_value_locked(uaddr
, uval
, newval
);
1411 if (unlikely(curval
== -EFAULT
))
1413 if (unlikely(curval
!= uval
))
1417 * We took the lock due to owner died take over.
1419 if (unlikely(lock_taken
))
1420 goto out_unlock_release_sem
;
1423 * We dont have the lock. Look up the PI state (or create it if
1424 * we are the first waiter):
1426 ret
= lookup_pi_state(uval
, hb
, &q
.key
, &q
.pi_state
);
1428 if (unlikely(ret
)) {
1433 * Task is exiting and we just wait for the
1436 queue_unlock(&q
, hb
);
1442 * No owner found for this futex. Check if the
1443 * OWNER_DIED bit is set to figure out whether
1444 * this is a robust futex or not.
1446 if (get_futex_value_locked(&curval
, uaddr
))
1450 * We simply start over in case of a robust
1451 * futex. The code above will take the futex
1454 if (curval
& FUTEX_OWNER_DIED
) {
1459 goto out_unlock_release_sem
;
1464 * Only actually queue now that the atomic ops are done:
1468 WARN_ON(!q
.pi_state
);
1470 * Block on the PI mutex:
1473 ret
= rt_mutex_timed_lock(&q
.pi_state
->pi_mutex
, to
, 1);
1475 ret
= rt_mutex_trylock(&q
.pi_state
->pi_mutex
);
1476 /* Fixup the trylock return value: */
1477 ret
= ret
? 0 : -EWOULDBLOCK
;
1480 spin_lock(q
.lock_ptr
);
1484 * Got the lock. We might not be the anticipated owner
1485 * if we did a lock-steal - fix up the PI-state in
1488 if (q
.pi_state
->owner
!= curr
)
1489 ret
= fixup_pi_state_owner(uaddr
, &q
, curr
, fshared
);
1492 * Catch the rare case, where the lock was released
1493 * when we were on the way back before we locked the
1496 if (q
.pi_state
->owner
== curr
) {
1498 * Try to get the rt_mutex now. This might
1499 * fail as some other task acquired the
1500 * rt_mutex after we removed ourself from the
1501 * rt_mutex waiters list.
1503 if (rt_mutex_trylock(&q
.pi_state
->pi_mutex
))
1507 * pi_state is incorrect, some other
1508 * task did a lock steal and we
1509 * returned due to timeout or signal
1510 * without taking the rt_mutex. Too
1511 * late. We can access the
1512 * rt_mutex_owner without locking, as
1513 * the other task is now blocked on
1514 * the hash bucket lock. Fix the state
1517 struct task_struct
*owner
;
1520 owner
= rt_mutex_owner(&q
.pi_state
->pi_mutex
);
1521 res
= fixup_pi_state_owner(uaddr
, &q
, owner
,
1524 /* propagate -EFAULT, if the fixup failed */
1530 * Paranoia check. If we did not take the lock
1531 * in the trylock above, then we should not be
1532 * the owner of the rtmutex, neither the real
1533 * nor the pending one:
1535 if (rt_mutex_owner(&q
.pi_state
->pi_mutex
) == curr
)
1536 printk(KERN_ERR
"futex_lock_pi: ret = %d "
1537 "pi-mutex: %p pi-state %p\n", ret
,
1538 q
.pi_state
->pi_mutex
.owner
,
1543 /* Unqueue and drop the lock */
1547 destroy_hrtimer_on_stack(&to
->timer
);
1548 return ret
!= -EINTR
? ret
: -ERESTARTNOINTR
;
1550 out_unlock_release_sem
:
1551 queue_unlock(&q
, hb
);
1554 put_futex_key(fshared
, &q
.key
);
1556 destroy_hrtimer_on_stack(&to
->timer
);
1561 * We have to r/w *(int __user *)uaddr, but we can't modify it
1562 * non-atomically. Therefore, if get_user below is not
1563 * enough, we need to handle the fault ourselves, while
1564 * still holding the mmap_sem.
1566 * ... and hb->lock. :-) --ANK
1568 queue_unlock(&q
, hb
);
1571 ret
= futex_handle_fault((unsigned long)uaddr
, attempt
);
1573 goto out_release_sem
;
1574 goto retry_unlocked
;
1577 ret
= get_user(uval
, uaddr
);
1578 if (!ret
&& (uval
!= -EFAULT
))
1582 destroy_hrtimer_on_stack(&to
->timer
);
1587 * Userspace attempted a TID -> 0 atomic transition, and failed.
1588 * This is the in-kernel slowpath: we look up the PI state (if any),
1589 * and do the rt-mutex unlock.
1591 static int futex_unlock_pi(u32 __user
*uaddr
, int fshared
)
1593 struct futex_hash_bucket
*hb
;
1594 struct futex_q
*this, *next
;
1596 struct plist_head
*head
;
1597 union futex_key key
= FUTEX_KEY_INIT
;
1598 int ret
, attempt
= 0;
1601 if (get_user(uval
, uaddr
))
1604 * We release only a lock we actually own:
1606 if ((uval
& FUTEX_TID_MASK
) != task_pid_vnr(current
))
1609 ret
= get_futex_key(uaddr
, fshared
, &key
);
1610 if (unlikely(ret
!= 0))
1613 hb
= hash_futex(&key
);
1615 spin_lock(&hb
->lock
);
1618 * To avoid races, try to do the TID -> 0 atomic transition
1619 * again. If it succeeds then we can return without waking
1622 if (!(uval
& FUTEX_OWNER_DIED
))
1623 uval
= cmpxchg_futex_value_locked(uaddr
, task_pid_vnr(current
), 0);
1626 if (unlikely(uval
== -EFAULT
))
1629 * Rare case: we managed to release the lock atomically,
1630 * no need to wake anyone else up:
1632 if (unlikely(uval
== task_pid_vnr(current
)))
1636 * Ok, other tasks may need to be woken up - check waiters
1637 * and do the wakeup if necessary:
1641 plist_for_each_entry_safe(this, next
, head
, list
) {
1642 if (!match_futex (&this->key
, &key
))
1644 ret
= wake_futex_pi(uaddr
, uval
, this);
1646 * The atomic access to the futex value
1647 * generated a pagefault, so retry the
1648 * user-access and the wakeup:
1655 * No waiters - kernel unlocks the futex:
1657 if (!(uval
& FUTEX_OWNER_DIED
)) {
1658 ret
= unlock_futex_pi(uaddr
, uval
);
1664 spin_unlock(&hb
->lock
);
1666 put_futex_key(fshared
, &key
);
1672 * We have to r/w *(int __user *)uaddr, but we can't modify it
1673 * non-atomically. Therefore, if get_user below is not
1674 * enough, we need to handle the fault ourselves, while
1675 * still holding the mmap_sem.
1677 * ... and hb->lock. --ANK
1679 spin_unlock(&hb
->lock
);
1682 ret
= futex_handle_fault((unsigned long)uaddr
, attempt
);
1686 goto retry_unlocked
;
1689 ret
= get_user(uval
, uaddr
);
1690 if (!ret
&& (uval
!= -EFAULT
))
1697 * Support for robust futexes: the kernel cleans up held futexes at
1700 * Implementation: user-space maintains a per-thread list of locks it
1701 * is holding. Upon do_exit(), the kernel carefully walks this list,
1702 * and marks all locks that are owned by this thread with the
1703 * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is
1704 * always manipulated with the lock held, so the list is private and
1705 * per-thread. Userspace also maintains a per-thread 'list_op_pending'
1706 * field, to allow the kernel to clean up if the thread dies after
1707 * acquiring the lock, but just before it could have added itself to
1708 * the list. There can only be one such pending lock.
1712 * sys_set_robust_list - set the robust-futex list head of a task
1713 * @head: pointer to the list-head
1714 * @len: length of the list-head, as userspace expects
1717 sys_set_robust_list(struct robust_list_head __user
*head
,
1720 if (!futex_cmpxchg_enabled
)
1723 * The kernel knows only one size for now:
1725 if (unlikely(len
!= sizeof(*head
)))
1728 current
->robust_list
= head
;
1734 * sys_get_robust_list - get the robust-futex list head of a task
1735 * @pid: pid of the process [zero for current task]
1736 * @head_ptr: pointer to a list-head pointer, the kernel fills it in
1737 * @len_ptr: pointer to a length field, the kernel fills in the header size
1740 sys_get_robust_list(int pid
, struct robust_list_head __user
* __user
*head_ptr
,
1741 size_t __user
*len_ptr
)
1743 struct robust_list_head __user
*head
;
1746 if (!futex_cmpxchg_enabled
)
1750 head
= current
->robust_list
;
1752 struct task_struct
*p
;
1756 p
= find_task_by_vpid(pid
);
1760 if ((current
->euid
!= p
->euid
) && (current
->euid
!= p
->uid
) &&
1761 !capable(CAP_SYS_PTRACE
))
1763 head
= p
->robust_list
;
1767 if (put_user(sizeof(*head
), len_ptr
))
1769 return put_user(head
, head_ptr
);
1778 * Process a futex-list entry, check whether it's owned by the
1779 * dying task, and do notification if so:
1781 int handle_futex_death(u32 __user
*uaddr
, struct task_struct
*curr
, int pi
)
1783 u32 uval
, nval
, mval
;
1786 if (get_user(uval
, uaddr
))
1789 if ((uval
& FUTEX_TID_MASK
) == task_pid_vnr(curr
)) {
1791 * Ok, this dying thread is truly holding a futex
1792 * of interest. Set the OWNER_DIED bit atomically
1793 * via cmpxchg, and if the value had FUTEX_WAITERS
1794 * set, wake up a waiter (if any). (We have to do a
1795 * futex_wake() even if OWNER_DIED is already set -
1796 * to handle the rare but possible case of recursive
1797 * thread-death.) The rest of the cleanup is done in
1800 mval
= (uval
& FUTEX_WAITERS
) | FUTEX_OWNER_DIED
;
1801 nval
= futex_atomic_cmpxchg_inatomic(uaddr
, uval
, mval
);
1803 if (nval
== -EFAULT
)
1810 * Wake robust non-PI futexes here. The wakeup of
1811 * PI futexes happens in exit_pi_state():
1813 if (!pi
&& (uval
& FUTEX_WAITERS
))
1814 futex_wake(uaddr
, 1, 1, FUTEX_BITSET_MATCH_ANY
);
1820 * Fetch a robust-list pointer. Bit 0 signals PI futexes:
1822 static inline int fetch_robust_entry(struct robust_list __user
**entry
,
1823 struct robust_list __user
* __user
*head
,
1826 unsigned long uentry
;
1828 if (get_user(uentry
, (unsigned long __user
*)head
))
1831 *entry
= (void __user
*)(uentry
& ~1UL);
1838 * Walk curr->robust_list (very carefully, it's a userspace list!)
1839 * and mark any locks found there dead, and notify any waiters.
1841 * We silently return on any sign of list-walking problem.
1843 void exit_robust_list(struct task_struct
*curr
)
1845 struct robust_list_head __user
*head
= curr
->robust_list
;
1846 struct robust_list __user
*entry
, *next_entry
, *pending
;
1847 unsigned int limit
= ROBUST_LIST_LIMIT
, pi
, next_pi
, pip
;
1848 unsigned long futex_offset
;
1851 if (!futex_cmpxchg_enabled
)
1855 * Fetch the list head (which was registered earlier, via
1856 * sys_set_robust_list()):
1858 if (fetch_robust_entry(&entry
, &head
->list
.next
, &pi
))
1861 * Fetch the relative futex offset:
1863 if (get_user(futex_offset
, &head
->futex_offset
))
1866 * Fetch any possibly pending lock-add first, and handle it
1869 if (fetch_robust_entry(&pending
, &head
->list_op_pending
, &pip
))
1872 next_entry
= NULL
; /* avoid warning with gcc */
1873 while (entry
!= &head
->list
) {
1875 * Fetch the next entry in the list before calling
1876 * handle_futex_death:
1878 rc
= fetch_robust_entry(&next_entry
, &entry
->next
, &next_pi
);
1880 * A pending lock might already be on the list, so
1881 * don't process it twice:
1883 if (entry
!= pending
)
1884 if (handle_futex_death((void __user
*)entry
+ futex_offset
,
1892 * Avoid excessively long or circular lists:
1901 handle_futex_death((void __user
*)pending
+ futex_offset
,
1905 long do_futex(u32 __user
*uaddr
, int op
, u32 val
, ktime_t
*timeout
,
1906 u32 __user
*uaddr2
, u32 val2
, u32 val3
)
1909 int cmd
= op
& FUTEX_CMD_MASK
;
1912 if (!(op
& FUTEX_PRIVATE_FLAG
))
1917 val3
= FUTEX_BITSET_MATCH_ANY
;
1918 case FUTEX_WAIT_BITSET
:
1919 ret
= futex_wait(uaddr
, fshared
, val
, timeout
, val3
);
1922 val3
= FUTEX_BITSET_MATCH_ANY
;
1923 case FUTEX_WAKE_BITSET
:
1924 ret
= futex_wake(uaddr
, fshared
, val
, val3
);
1927 ret
= futex_requeue(uaddr
, fshared
, uaddr2
, val
, val2
, NULL
);
1929 case FUTEX_CMP_REQUEUE
:
1930 ret
= futex_requeue(uaddr
, fshared
, uaddr2
, val
, val2
, &val3
);
1933 ret
= futex_wake_op(uaddr
, fshared
, uaddr2
, val
, val2
, val3
);
1936 if (futex_cmpxchg_enabled
)
1937 ret
= futex_lock_pi(uaddr
, fshared
, val
, timeout
, 0);
1939 case FUTEX_UNLOCK_PI
:
1940 if (futex_cmpxchg_enabled
)
1941 ret
= futex_unlock_pi(uaddr
, fshared
);
1943 case FUTEX_TRYLOCK_PI
:
1944 if (futex_cmpxchg_enabled
)
1945 ret
= futex_lock_pi(uaddr
, fshared
, 0, timeout
, 1);
1954 asmlinkage
long sys_futex(u32 __user
*uaddr
, int op
, u32 val
,
1955 struct timespec __user
*utime
, u32 __user
*uaddr2
,
1959 ktime_t t
, *tp
= NULL
;
1961 int cmd
= op
& FUTEX_CMD_MASK
;
1963 if (utime
&& (cmd
== FUTEX_WAIT
|| cmd
== FUTEX_LOCK_PI
||
1964 cmd
== FUTEX_WAIT_BITSET
)) {
1965 if (copy_from_user(&ts
, utime
, sizeof(ts
)) != 0)
1967 if (!timespec_valid(&ts
))
1970 t
= timespec_to_ktime(ts
);
1971 if (cmd
== FUTEX_WAIT
)
1972 t
= ktime_add_safe(ktime_get(), t
);
1976 * requeue parameter in 'utime' if cmd == FUTEX_REQUEUE.
1977 * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP.
1979 if (cmd
== FUTEX_REQUEUE
|| cmd
== FUTEX_CMP_REQUEUE
||
1980 cmd
== FUTEX_WAKE_OP
)
1981 val2
= (u32
) (unsigned long) utime
;
1983 return do_futex(uaddr
, op
, val
, tp
, uaddr2
, val2
, val3
);
1986 static int __init
futex_init(void)
1992 * This will fail and we want it. Some arch implementations do
1993 * runtime detection of the futex_atomic_cmpxchg_inatomic()
1994 * functionality. We want to know that before we call in any
1995 * of the complex code paths. Also we want to prevent
1996 * registration of robust lists in that case. NULL is
1997 * guaranteed to fault and we get -EFAULT on functional
1998 * implementation, the non functional ones will return
2001 curval
= cmpxchg_futex_value_locked(NULL
, 0, 0);
2002 if (curval
== -EFAULT
)
2003 futex_cmpxchg_enabled
= 1;
2005 for (i
= 0; i
< ARRAY_SIZE(futex_queues
); i
++) {
2006 plist_head_init(&futex_queues
[i
].chain
, &futex_queues
[i
].lock
);
2007 spin_lock_init(&futex_queues
[i
].lock
);
2012 __initcall(futex_init
);