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 * Requeue-PI support by Darren Hart <dvhltc@us.ibm.com>
23 * Copyright (C) IBM Corporation, 2009
24 * Thanks to Thomas Gleixner for conceptual design and careful reviews.
26 * Thanks to Ben LaHaise for yelling "hashed waitqueues" loudly
27 * enough at me, Linus for the original (flawed) idea, Matthew
28 * Kirkwood for proof-of-concept implementation.
30 * "The futexes are also cursed."
31 * "But they come in a choice of three flavours!"
33 * This program is free software; you can redistribute it and/or modify
34 * it under the terms of the GNU General Public License as published by
35 * the Free Software Foundation; either version 2 of the License, or
36 * (at your option) any later version.
38 * This program is distributed in the hope that it will be useful,
39 * but WITHOUT ANY WARRANTY; without even the implied warranty of
40 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
41 * GNU General Public License for more details.
43 * You should have received a copy of the GNU General Public License
44 * along with this program; if not, write to the Free Software
45 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
47 #include <linux/slab.h>
48 #include <linux/poll.h>
50 #include <linux/file.h>
51 #include <linux/jhash.h>
52 #include <linux/init.h>
53 #include <linux/futex.h>
54 #include <linux/mount.h>
55 #include <linux/pagemap.h>
56 #include <linux/syscalls.h>
57 #include <linux/signal.h>
58 #include <linux/module.h>
59 #include <linux/magic.h>
60 #include <linux/pid.h>
61 #include <linux/nsproxy.h>
63 #include <asm/futex.h>
65 #include "rtmutex_common.h"
67 int __read_mostly futex_cmpxchg_enabled
;
69 #define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8)
72 * Priority Inheritance state:
74 struct futex_pi_state
{
76 * list of 'owned' pi_state instances - these have to be
77 * cleaned up in do_exit() if the task exits prematurely:
79 struct list_head list
;
84 struct rt_mutex pi_mutex
;
86 struct task_struct
*owner
;
93 * We use this hashed waitqueue instead of a normal wait_queue_t, so
94 * we can wake only the relevant ones (hashed queues may be shared).
96 * A futex_q has a woken state, just like tasks have TASK_RUNNING.
97 * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0.
98 * The order of wakup is always to make the first condition true, then
99 * wake up q->waiter, then make the second condition true.
102 struct plist_node list
;
103 /* Waiter reference */
104 struct task_struct
*task
;
106 /* Which hash list lock to use: */
107 spinlock_t
*lock_ptr
;
109 /* Key which the futex is hashed on: */
112 /* Optional priority inheritance state: */
113 struct futex_pi_state
*pi_state
;
115 /* rt_waiter storage for requeue_pi: */
116 struct rt_mutex_waiter
*rt_waiter
;
118 /* The expected requeue pi target futex key: */
119 union futex_key
*requeue_pi_key
;
121 /* Bitset for the optional bitmasked wakeup */
126 * Hash buckets are shared by all the futex_keys that hash to the same
127 * location. Each key may have multiple futex_q structures, one for each task
128 * waiting on a futex.
130 struct futex_hash_bucket
{
132 struct plist_head chain
;
135 static struct futex_hash_bucket futex_queues
[1<<FUTEX_HASHBITS
];
138 * We hash on the keys returned from get_futex_key (see below).
140 static struct futex_hash_bucket
*hash_futex(union futex_key
*key
)
142 u32 hash
= jhash2((u32
*)&key
->both
.word
,
143 (sizeof(key
->both
.word
)+sizeof(key
->both
.ptr
))/4,
145 return &futex_queues
[hash
& ((1 << FUTEX_HASHBITS
)-1)];
149 * Return 1 if two futex_keys are equal, 0 otherwise.
151 static inline int match_futex(union futex_key
*key1
, union futex_key
*key2
)
154 && key1
->both
.word
== key2
->both
.word
155 && key1
->both
.ptr
== key2
->both
.ptr
156 && key1
->both
.offset
== key2
->both
.offset
);
160 * Take a reference to the resource addressed by a key.
161 * Can be called while holding spinlocks.
164 static void get_futex_key_refs(union futex_key
*key
)
169 switch (key
->both
.offset
& (FUT_OFF_INODE
|FUT_OFF_MMSHARED
)) {
171 atomic_inc(&key
->shared
.inode
->i_count
);
173 case FUT_OFF_MMSHARED
:
174 atomic_inc(&key
->private.mm
->mm_count
);
180 * Drop a reference to the resource addressed by a key.
181 * The hash bucket spinlock must not be held.
183 static void drop_futex_key_refs(union futex_key
*key
)
185 if (!key
->both
.ptr
) {
186 /* If we're here then we tried to put a key we failed to get */
191 switch (key
->both
.offset
& (FUT_OFF_INODE
|FUT_OFF_MMSHARED
)) {
193 iput(key
->shared
.inode
);
195 case FUT_OFF_MMSHARED
:
196 mmdrop(key
->private.mm
);
202 * get_futex_key - Get parameters which are the keys for a futex.
203 * @uaddr: virtual address of the futex
204 * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED
205 * @key: address where result is stored.
206 * @rw: mapping needs to be read/write (values: VERIFY_READ, VERIFY_WRITE)
208 * Returns a negative error code or 0
209 * The key words are stored in *key on success.
211 * For shared mappings, it's (page->index, vma->vm_file->f_path.dentry->d_inode,
212 * offset_within_page). For private mappings, it's (uaddr, current->mm).
213 * We can usually work out the index without swapping in the page.
215 * lock_page() might sleep, the caller should not hold a spinlock.
218 get_futex_key(u32 __user
*uaddr
, int fshared
, union futex_key
*key
, int rw
)
220 unsigned long address
= (unsigned long)uaddr
;
221 struct mm_struct
*mm
= current
->mm
;
226 * The futex address must be "naturally" aligned.
228 key
->both
.offset
= address
% PAGE_SIZE
;
229 if (unlikely((address
% sizeof(u32
)) != 0))
231 address
-= key
->both
.offset
;
234 * PROCESS_PRIVATE futexes are fast.
235 * As the mm cannot disappear under us and the 'key' only needs
236 * virtual address, we dont even have to find the underlying vma.
237 * Note : We do have to check 'uaddr' is a valid user address,
238 * but access_ok() should be faster than find_vma()
241 if (unlikely(!access_ok(rw
, uaddr
, sizeof(u32
))))
243 key
->private.mm
= mm
;
244 key
->private.address
= address
;
245 get_futex_key_refs(key
);
250 err
= get_user_pages_fast(address
, 1, rw
== VERIFY_WRITE
, &page
);
254 page
= compound_head(page
);
256 if (!page
->mapping
) {
263 * Private mappings are handled in a simple way.
265 * NOTE: When userspace waits on a MAP_SHARED mapping, even if
266 * it's a read-only handle, it's expected that futexes attach to
267 * the object not the particular process.
269 if (PageAnon(page
)) {
270 key
->both
.offset
|= FUT_OFF_MMSHARED
; /* ref taken on mm */
271 key
->private.mm
= mm
;
272 key
->private.address
= address
;
274 key
->both
.offset
|= FUT_OFF_INODE
; /* inode-based key */
275 key
->shared
.inode
= page
->mapping
->host
;
276 key
->shared
.pgoff
= page
->index
;
279 get_futex_key_refs(key
);
287 void put_futex_key(int fshared
, union futex_key
*key
)
289 drop_futex_key_refs(key
);
293 * fault_in_user_writeable - fault in user address and verify RW access
294 * @uaddr: pointer to faulting user space address
296 * Slow path to fixup the fault we just took in the atomic write
299 * We have no generic implementation of a non destructive write to the
300 * user address. We know that we faulted in the atomic pagefault
301 * disabled section so we can as well avoid the #PF overhead by
302 * calling get_user_pages() right away.
304 static int fault_in_user_writeable(u32 __user
*uaddr
)
306 struct mm_struct
*mm
= current
->mm
;
309 down_read(&mm
->mmap_sem
);
310 ret
= get_user_pages(current
, mm
, (unsigned long)uaddr
,
311 1, 1, 0, NULL
, NULL
);
312 up_read(&mm
->mmap_sem
);
314 return ret
< 0 ? ret
: 0;
318 * futex_top_waiter() - Return the highest priority waiter on a futex
319 * @hb: the hash bucket the futex_q's reside in
320 * @key: the futex key (to distinguish it from other futex futex_q's)
322 * Must be called with the hb lock held.
324 static struct futex_q
*futex_top_waiter(struct futex_hash_bucket
*hb
,
325 union futex_key
*key
)
327 struct futex_q
*this;
329 plist_for_each_entry(this, &hb
->chain
, list
) {
330 if (match_futex(&this->key
, key
))
336 static u32
cmpxchg_futex_value_locked(u32 __user
*uaddr
, u32 uval
, u32 newval
)
341 curval
= futex_atomic_cmpxchg_inatomic(uaddr
, uval
, newval
);
347 static int get_futex_value_locked(u32
*dest
, u32 __user
*from
)
352 ret
= __copy_from_user_inatomic(dest
, from
, sizeof(u32
));
355 return ret
? -EFAULT
: 0;
362 static int refill_pi_state_cache(void)
364 struct futex_pi_state
*pi_state
;
366 if (likely(current
->pi_state_cache
))
369 pi_state
= kzalloc(sizeof(*pi_state
), GFP_KERNEL
);
374 INIT_LIST_HEAD(&pi_state
->list
);
375 /* pi_mutex gets initialized later */
376 pi_state
->owner
= NULL
;
377 atomic_set(&pi_state
->refcount
, 1);
378 pi_state
->key
= FUTEX_KEY_INIT
;
380 current
->pi_state_cache
= pi_state
;
385 static struct futex_pi_state
* alloc_pi_state(void)
387 struct futex_pi_state
*pi_state
= current
->pi_state_cache
;
390 current
->pi_state_cache
= NULL
;
395 static void free_pi_state(struct futex_pi_state
*pi_state
)
397 if (!atomic_dec_and_test(&pi_state
->refcount
))
401 * If pi_state->owner is NULL, the owner is most probably dying
402 * and has cleaned up the pi_state already
404 if (pi_state
->owner
) {
405 spin_lock_irq(&pi_state
->owner
->pi_lock
);
406 list_del_init(&pi_state
->list
);
407 spin_unlock_irq(&pi_state
->owner
->pi_lock
);
409 rt_mutex_proxy_unlock(&pi_state
->pi_mutex
, pi_state
->owner
);
412 if (current
->pi_state_cache
)
416 * pi_state->list is already empty.
417 * clear pi_state->owner.
418 * refcount is at 0 - put it back to 1.
420 pi_state
->owner
= NULL
;
421 atomic_set(&pi_state
->refcount
, 1);
422 current
->pi_state_cache
= pi_state
;
427 * Look up the task based on what TID userspace gave us.
430 static struct task_struct
* futex_find_get_task(pid_t pid
)
432 struct task_struct
*p
;
433 const struct cred
*cred
= current_cred(), *pcred
;
436 p
= find_task_by_vpid(pid
);
440 pcred
= __task_cred(p
);
441 if (cred
->euid
!= pcred
->euid
&&
442 cred
->euid
!= pcred
->uid
)
454 * This task is holding PI mutexes at exit time => bad.
455 * Kernel cleans up PI-state, but userspace is likely hosed.
456 * (Robust-futex cleanup is separate and might save the day for userspace.)
458 void exit_pi_state_list(struct task_struct
*curr
)
460 struct list_head
*next
, *head
= &curr
->pi_state_list
;
461 struct futex_pi_state
*pi_state
;
462 struct futex_hash_bucket
*hb
;
463 union futex_key key
= FUTEX_KEY_INIT
;
465 if (!futex_cmpxchg_enabled
)
468 * We are a ZOMBIE and nobody can enqueue itself on
469 * pi_state_list anymore, but we have to be careful
470 * versus waiters unqueueing themselves:
472 spin_lock_irq(&curr
->pi_lock
);
473 while (!list_empty(head
)) {
476 pi_state
= list_entry(next
, struct futex_pi_state
, list
);
478 hb
= hash_futex(&key
);
479 spin_unlock_irq(&curr
->pi_lock
);
481 spin_lock(&hb
->lock
);
483 spin_lock_irq(&curr
->pi_lock
);
485 * We dropped the pi-lock, so re-check whether this
486 * task still owns the PI-state:
488 if (head
->next
!= next
) {
489 spin_unlock(&hb
->lock
);
493 WARN_ON(pi_state
->owner
!= curr
);
494 WARN_ON(list_empty(&pi_state
->list
));
495 list_del_init(&pi_state
->list
);
496 pi_state
->owner
= NULL
;
497 spin_unlock_irq(&curr
->pi_lock
);
499 rt_mutex_unlock(&pi_state
->pi_mutex
);
501 spin_unlock(&hb
->lock
);
503 spin_lock_irq(&curr
->pi_lock
);
505 spin_unlock_irq(&curr
->pi_lock
);
509 lookup_pi_state(u32 uval
, struct futex_hash_bucket
*hb
,
510 union futex_key
*key
, struct futex_pi_state
**ps
)
512 struct futex_pi_state
*pi_state
= NULL
;
513 struct futex_q
*this, *next
;
514 struct plist_head
*head
;
515 struct task_struct
*p
;
516 pid_t pid
= uval
& FUTEX_TID_MASK
;
520 plist_for_each_entry_safe(this, next
, head
, list
) {
521 if (match_futex(&this->key
, key
)) {
523 * Another waiter already exists - bump up
524 * the refcount and return its pi_state:
526 pi_state
= this->pi_state
;
528 * Userspace might have messed up non PI and PI futexes
530 if (unlikely(!pi_state
))
533 WARN_ON(!atomic_read(&pi_state
->refcount
));
536 * When pi_state->owner is NULL then the owner died
537 * and another waiter is on the fly. pi_state->owner
538 * is fixed up by the task which acquires
539 * pi_state->rt_mutex.
541 * We do not check for pid == 0 which can happen when
542 * the owner died and robust_list_exit() cleared the
545 if (pid
&& pi_state
->owner
) {
547 * Bail out if user space manipulated the
550 if (pid
!= task_pid_vnr(pi_state
->owner
))
554 atomic_inc(&pi_state
->refcount
);
562 * We are the first waiter - try to look up the real owner and attach
563 * the new pi_state to it, but bail out when TID = 0
567 p
= futex_find_get_task(pid
);
572 * We need to look at the task state flags to figure out,
573 * whether the task is exiting. To protect against the do_exit
574 * change of the task flags, we do this protected by
577 spin_lock_irq(&p
->pi_lock
);
578 if (unlikely(p
->flags
& PF_EXITING
)) {
580 * The task is on the way out. When PF_EXITPIDONE is
581 * set, we know that the task has finished the
584 int ret
= (p
->flags
& PF_EXITPIDONE
) ? -ESRCH
: -EAGAIN
;
586 spin_unlock_irq(&p
->pi_lock
);
591 pi_state
= alloc_pi_state();
594 * Initialize the pi_mutex in locked state and make 'p'
597 rt_mutex_init_proxy_locked(&pi_state
->pi_mutex
, p
);
599 /* Store the key for possible exit cleanups: */
600 pi_state
->key
= *key
;
602 WARN_ON(!list_empty(&pi_state
->list
));
603 list_add(&pi_state
->list
, &p
->pi_state_list
);
605 spin_unlock_irq(&p
->pi_lock
);
615 * futex_lock_pi_atomic() - atomic work required to acquire a pi aware futex
616 * @uaddr: the pi futex user address
617 * @hb: the pi futex hash bucket
618 * @key: the futex key associated with uaddr and hb
619 * @ps: the pi_state pointer where we store the result of the
621 * @task: the task to perform the atomic lock work for. This will
622 * be "current" except in the case of requeue pi.
623 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
627 * 1 - acquired the lock
630 * The hb->lock and futex_key refs shall be held by the caller.
632 static int futex_lock_pi_atomic(u32 __user
*uaddr
, struct futex_hash_bucket
*hb
,
633 union futex_key
*key
,
634 struct futex_pi_state
**ps
,
635 struct task_struct
*task
, int set_waiters
)
637 int lock_taken
, ret
, ownerdied
= 0;
638 u32 uval
, newval
, curval
;
641 ret
= lock_taken
= 0;
644 * To avoid races, we attempt to take the lock here again
645 * (by doing a 0 -> TID atomic cmpxchg), while holding all
646 * the locks. It will most likely not succeed.
648 newval
= task_pid_vnr(task
);
650 newval
|= FUTEX_WAITERS
;
652 curval
= cmpxchg_futex_value_locked(uaddr
, 0, newval
);
654 if (unlikely(curval
== -EFAULT
))
660 if ((unlikely((curval
& FUTEX_TID_MASK
) == task_pid_vnr(task
))))
664 * Surprise - we got the lock. Just return to userspace:
666 if (unlikely(!curval
))
672 * Set the FUTEX_WAITERS flag, so the owner will know it has someone
673 * to wake at the next unlock.
675 newval
= curval
| FUTEX_WAITERS
;
678 * There are two cases, where a futex might have no owner (the
679 * owner TID is 0): OWNER_DIED. We take over the futex in this
680 * case. We also do an unconditional take over, when the owner
683 * This is safe as we are protected by the hash bucket lock !
685 if (unlikely(ownerdied
|| !(curval
& FUTEX_TID_MASK
))) {
686 /* Keep the OWNER_DIED bit */
687 newval
= (curval
& ~FUTEX_TID_MASK
) | task_pid_vnr(task
);
692 curval
= cmpxchg_futex_value_locked(uaddr
, uval
, newval
);
694 if (unlikely(curval
== -EFAULT
))
696 if (unlikely(curval
!= uval
))
700 * We took the lock due to owner died take over.
702 if (unlikely(lock_taken
))
706 * We dont have the lock. Look up the PI state (or create it if
707 * we are the first waiter):
709 ret
= lookup_pi_state(uval
, hb
, key
, ps
);
715 * No owner found for this futex. Check if the
716 * OWNER_DIED bit is set to figure out whether
717 * this is a robust futex or not.
719 if (get_futex_value_locked(&curval
, uaddr
))
723 * We simply start over in case of a robust
724 * futex. The code above will take the futex
727 if (curval
& FUTEX_OWNER_DIED
) {
740 * The hash bucket lock must be held when this is called.
741 * Afterwards, the futex_q must not be accessed.
743 static void wake_futex(struct futex_q
*q
)
745 struct task_struct
*p
= q
->task
;
748 * We set q->lock_ptr = NULL _before_ we wake up the task. If
749 * a non futex wake up happens on another CPU then the task
750 * might exit and p would dereference a non existing task
751 * struct. Prevent this by holding a reference on p across the
756 plist_del(&q
->list
, &q
->list
.plist
);
758 * The waiting task can free the futex_q as soon as
759 * q->lock_ptr = NULL is written, without taking any locks. A
760 * memory barrier is required here to prevent the following
761 * store to lock_ptr from getting ahead of the plist_del.
766 wake_up_state(p
, TASK_NORMAL
);
770 static int wake_futex_pi(u32 __user
*uaddr
, u32 uval
, struct futex_q
*this)
772 struct task_struct
*new_owner
;
773 struct futex_pi_state
*pi_state
= this->pi_state
;
780 * If current does not own the pi_state then the futex is
781 * inconsistent and user space fiddled with the futex value.
783 if (pi_state
->owner
!= current
)
786 spin_lock(&pi_state
->pi_mutex
.wait_lock
);
787 new_owner
= rt_mutex_next_owner(&pi_state
->pi_mutex
);
790 * This happens when we have stolen the lock and the original
791 * pending owner did not enqueue itself back on the rt_mutex.
792 * Thats not a tragedy. We know that way, that a lock waiter
793 * is on the fly. We make the futex_q waiter the pending owner.
796 new_owner
= this->task
;
799 * We pass it to the next owner. (The WAITERS bit is always
800 * kept enabled while there is PI state around. We must also
801 * preserve the owner died bit.)
803 if (!(uval
& FUTEX_OWNER_DIED
)) {
806 newval
= FUTEX_WAITERS
| task_pid_vnr(new_owner
);
808 curval
= cmpxchg_futex_value_locked(uaddr
, uval
, newval
);
810 if (curval
== -EFAULT
)
812 else if (curval
!= uval
)
815 spin_unlock(&pi_state
->pi_mutex
.wait_lock
);
820 spin_lock_irq(&pi_state
->owner
->pi_lock
);
821 WARN_ON(list_empty(&pi_state
->list
));
822 list_del_init(&pi_state
->list
);
823 spin_unlock_irq(&pi_state
->owner
->pi_lock
);
825 spin_lock_irq(&new_owner
->pi_lock
);
826 WARN_ON(!list_empty(&pi_state
->list
));
827 list_add(&pi_state
->list
, &new_owner
->pi_state_list
);
828 pi_state
->owner
= new_owner
;
829 spin_unlock_irq(&new_owner
->pi_lock
);
831 spin_unlock(&pi_state
->pi_mutex
.wait_lock
);
832 rt_mutex_unlock(&pi_state
->pi_mutex
);
837 static int unlock_futex_pi(u32 __user
*uaddr
, u32 uval
)
842 * There is no waiter, so we unlock the futex. The owner died
843 * bit has not to be preserved here. We are the owner:
845 oldval
= cmpxchg_futex_value_locked(uaddr
, uval
, 0);
847 if (oldval
== -EFAULT
)
856 * Express the locking dependencies for lockdep:
859 double_lock_hb(struct futex_hash_bucket
*hb1
, struct futex_hash_bucket
*hb2
)
862 spin_lock(&hb1
->lock
);
864 spin_lock_nested(&hb2
->lock
, SINGLE_DEPTH_NESTING
);
865 } else { /* hb1 > hb2 */
866 spin_lock(&hb2
->lock
);
867 spin_lock_nested(&hb1
->lock
, SINGLE_DEPTH_NESTING
);
872 double_unlock_hb(struct futex_hash_bucket
*hb1
, struct futex_hash_bucket
*hb2
)
874 spin_unlock(&hb1
->lock
);
876 spin_unlock(&hb2
->lock
);
880 * Wake up waiters matching bitset queued on this futex (uaddr).
882 static int futex_wake(u32 __user
*uaddr
, int fshared
, int nr_wake
, u32 bitset
)
884 struct futex_hash_bucket
*hb
;
885 struct futex_q
*this, *next
;
886 struct plist_head
*head
;
887 union futex_key key
= FUTEX_KEY_INIT
;
893 ret
= get_futex_key(uaddr
, fshared
, &key
, VERIFY_READ
);
894 if (unlikely(ret
!= 0))
897 hb
= hash_futex(&key
);
898 spin_lock(&hb
->lock
);
901 plist_for_each_entry_safe(this, next
, head
, list
) {
902 if (match_futex (&this->key
, &key
)) {
903 if (this->pi_state
|| this->rt_waiter
) {
908 /* Check if one of the bits is set in both bitsets */
909 if (!(this->bitset
& bitset
))
913 if (++ret
>= nr_wake
)
918 spin_unlock(&hb
->lock
);
919 put_futex_key(fshared
, &key
);
925 * Wake up all waiters hashed on the physical page that is mapped
926 * to this virtual address:
929 futex_wake_op(u32 __user
*uaddr1
, int fshared
, u32 __user
*uaddr2
,
930 int nr_wake
, int nr_wake2
, int op
)
932 union futex_key key1
= FUTEX_KEY_INIT
, key2
= FUTEX_KEY_INIT
;
933 struct futex_hash_bucket
*hb1
, *hb2
;
934 struct plist_head
*head
;
935 struct futex_q
*this, *next
;
939 ret
= get_futex_key(uaddr1
, fshared
, &key1
, VERIFY_READ
);
940 if (unlikely(ret
!= 0))
942 ret
= get_futex_key(uaddr2
, fshared
, &key2
, VERIFY_WRITE
);
943 if (unlikely(ret
!= 0))
946 hb1
= hash_futex(&key1
);
947 hb2
= hash_futex(&key2
);
950 double_lock_hb(hb1
, hb2
);
951 op_ret
= futex_atomic_op_inuser(op
, uaddr2
);
952 if (unlikely(op_ret
< 0)) {
954 double_unlock_hb(hb1
, hb2
);
958 * we don't get EFAULT from MMU faults if we don't have an MMU,
959 * but we might get them from range checking
965 if (unlikely(op_ret
!= -EFAULT
)) {
970 ret
= fault_in_user_writeable(uaddr2
);
977 put_futex_key(fshared
, &key2
);
978 put_futex_key(fshared
, &key1
);
984 plist_for_each_entry_safe(this, next
, head
, list
) {
985 if (match_futex (&this->key
, &key1
)) {
987 if (++ret
>= nr_wake
)
996 plist_for_each_entry_safe(this, next
, head
, list
) {
997 if (match_futex (&this->key
, &key2
)) {
999 if (++op_ret
>= nr_wake2
)
1006 double_unlock_hb(hb1
, hb2
);
1008 put_futex_key(fshared
, &key2
);
1010 put_futex_key(fshared
, &key1
);
1016 * requeue_futex() - Requeue a futex_q from one hb to another
1017 * @q: the futex_q to requeue
1018 * @hb1: the source hash_bucket
1019 * @hb2: the target hash_bucket
1020 * @key2: the new key for the requeued futex_q
1023 void requeue_futex(struct futex_q
*q
, struct futex_hash_bucket
*hb1
,
1024 struct futex_hash_bucket
*hb2
, union futex_key
*key2
)
1028 * If key1 and key2 hash to the same bucket, no need to
1031 if (likely(&hb1
->chain
!= &hb2
->chain
)) {
1032 plist_del(&q
->list
, &hb1
->chain
);
1033 plist_add(&q
->list
, &hb2
->chain
);
1034 q
->lock_ptr
= &hb2
->lock
;
1035 #ifdef CONFIG_DEBUG_PI_LIST
1036 q
->list
.plist
.lock
= &hb2
->lock
;
1039 get_futex_key_refs(key2
);
1044 * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue
1046 * key: the key of the requeue target futex
1047 * hb: the hash_bucket of the requeue target futex
1049 * During futex_requeue, with requeue_pi=1, it is possible to acquire the
1050 * target futex if it is uncontended or via a lock steal. Set the futex_q key
1051 * to the requeue target futex so the waiter can detect the wakeup on the right
1052 * futex, but remove it from the hb and NULL the rt_waiter so it can detect
1053 * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock
1054 * to protect access to the pi_state to fixup the owner later. Must be called
1055 * with both q->lock_ptr and hb->lock held.
1058 void requeue_pi_wake_futex(struct futex_q
*q
, union futex_key
*key
,
1059 struct futex_hash_bucket
*hb
)
1061 get_futex_key_refs(key
);
1064 WARN_ON(plist_node_empty(&q
->list
));
1065 plist_del(&q
->list
, &q
->list
.plist
);
1067 WARN_ON(!q
->rt_waiter
);
1068 q
->rt_waiter
= NULL
;
1070 q
->lock_ptr
= &hb
->lock
;
1071 #ifdef CONFIG_DEBUG_PI_LIST
1072 q
->list
.plist
.lock
= &hb
->lock
;
1075 wake_up_state(q
->task
, TASK_NORMAL
);
1079 * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter
1080 * @pifutex: the user address of the to futex
1081 * @hb1: the from futex hash bucket, must be locked by the caller
1082 * @hb2: the to futex hash bucket, must be locked by the caller
1083 * @key1: the from futex key
1084 * @key2: the to futex key
1085 * @ps: address to store the pi_state pointer
1086 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
1088 * Try and get the lock on behalf of the top waiter if we can do it atomically.
1089 * Wake the top waiter if we succeed. If the caller specified set_waiters,
1090 * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit.
1091 * hb1 and hb2 must be held by the caller.
1094 * 0 - failed to acquire the lock atomicly
1095 * 1 - acquired the lock
1098 static int futex_proxy_trylock_atomic(u32 __user
*pifutex
,
1099 struct futex_hash_bucket
*hb1
,
1100 struct futex_hash_bucket
*hb2
,
1101 union futex_key
*key1
, union futex_key
*key2
,
1102 struct futex_pi_state
**ps
, int set_waiters
)
1104 struct futex_q
*top_waiter
= NULL
;
1108 if (get_futex_value_locked(&curval
, pifutex
))
1112 * Find the top_waiter and determine if there are additional waiters.
1113 * If the caller intends to requeue more than 1 waiter to pifutex,
1114 * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now,
1115 * as we have means to handle the possible fault. If not, don't set
1116 * the bit unecessarily as it will force the subsequent unlock to enter
1119 top_waiter
= futex_top_waiter(hb1
, key1
);
1121 /* There are no waiters, nothing for us to do. */
1125 /* Ensure we requeue to the expected futex. */
1126 if (!match_futex(top_waiter
->requeue_pi_key
, key2
))
1130 * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in
1131 * the contended case or if set_waiters is 1. The pi_state is returned
1132 * in ps in contended cases.
1134 ret
= futex_lock_pi_atomic(pifutex
, hb2
, key2
, ps
, top_waiter
->task
,
1137 requeue_pi_wake_futex(top_waiter
, key2
, hb2
);
1143 * futex_requeue() - Requeue waiters from uaddr1 to uaddr2
1144 * uaddr1: source futex user address
1145 * uaddr2: target futex user address
1146 * nr_wake: number of waiters to wake (must be 1 for requeue_pi)
1147 * nr_requeue: number of waiters to requeue (0-INT_MAX)
1148 * requeue_pi: if we are attempting to requeue from a non-pi futex to a
1149 * pi futex (pi to pi requeue is not supported)
1151 * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire
1152 * uaddr2 atomically on behalf of the top waiter.
1155 * >=0 - on success, the number of tasks requeued or woken
1158 static int futex_requeue(u32 __user
*uaddr1
, int fshared
, u32 __user
*uaddr2
,
1159 int nr_wake
, int nr_requeue
, u32
*cmpval
,
1162 union futex_key key1
= FUTEX_KEY_INIT
, key2
= FUTEX_KEY_INIT
;
1163 int drop_count
= 0, task_count
= 0, ret
;
1164 struct futex_pi_state
*pi_state
= NULL
;
1165 struct futex_hash_bucket
*hb1
, *hb2
;
1166 struct plist_head
*head1
;
1167 struct futex_q
*this, *next
;
1172 * requeue_pi requires a pi_state, try to allocate it now
1173 * without any locks in case it fails.
1175 if (refill_pi_state_cache())
1178 * requeue_pi must wake as many tasks as it can, up to nr_wake
1179 * + nr_requeue, since it acquires the rt_mutex prior to
1180 * returning to userspace, so as to not leave the rt_mutex with
1181 * waiters and no owner. However, second and third wake-ups
1182 * cannot be predicted as they involve race conditions with the
1183 * first wake and a fault while looking up the pi_state. Both
1184 * pthread_cond_signal() and pthread_cond_broadcast() should
1192 if (pi_state
!= NULL
) {
1194 * We will have to lookup the pi_state again, so free this one
1195 * to keep the accounting correct.
1197 free_pi_state(pi_state
);
1201 ret
= get_futex_key(uaddr1
, fshared
, &key1
, VERIFY_READ
);
1202 if (unlikely(ret
!= 0))
1204 ret
= get_futex_key(uaddr2
, fshared
, &key2
,
1205 requeue_pi
? VERIFY_WRITE
: VERIFY_READ
);
1206 if (unlikely(ret
!= 0))
1209 hb1
= hash_futex(&key1
);
1210 hb2
= hash_futex(&key2
);
1213 double_lock_hb(hb1
, hb2
);
1215 if (likely(cmpval
!= NULL
)) {
1218 ret
= get_futex_value_locked(&curval
, uaddr1
);
1220 if (unlikely(ret
)) {
1221 double_unlock_hb(hb1
, hb2
);
1223 ret
= get_user(curval
, uaddr1
);
1230 put_futex_key(fshared
, &key2
);
1231 put_futex_key(fshared
, &key1
);
1234 if (curval
!= *cmpval
) {
1240 if (requeue_pi
&& (task_count
- nr_wake
< nr_requeue
)) {
1242 * Attempt to acquire uaddr2 and wake the top waiter. If we
1243 * intend to requeue waiters, force setting the FUTEX_WAITERS
1244 * bit. We force this here where we are able to easily handle
1245 * faults rather in the requeue loop below.
1247 ret
= futex_proxy_trylock_atomic(uaddr2
, hb1
, hb2
, &key1
,
1248 &key2
, &pi_state
, nr_requeue
);
1251 * At this point the top_waiter has either taken uaddr2 or is
1252 * waiting on it. If the former, then the pi_state will not
1253 * exist yet, look it up one more time to ensure we have a
1260 ret
= get_futex_value_locked(&curval2
, uaddr2
);
1262 ret
= lookup_pi_state(curval2
, hb2
, &key2
,
1270 double_unlock_hb(hb1
, hb2
);
1271 put_futex_key(fshared
, &key2
);
1272 put_futex_key(fshared
, &key1
);
1273 ret
= fault_in_user_writeable(uaddr2
);
1278 /* The owner was exiting, try again. */
1279 double_unlock_hb(hb1
, hb2
);
1280 put_futex_key(fshared
, &key2
);
1281 put_futex_key(fshared
, &key1
);
1289 head1
= &hb1
->chain
;
1290 plist_for_each_entry_safe(this, next
, head1
, list
) {
1291 if (task_count
- nr_wake
>= nr_requeue
)
1294 if (!match_futex(&this->key
, &key1
))
1298 * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always
1299 * be paired with each other and no other futex ops.
1301 if ((requeue_pi
&& !this->rt_waiter
) ||
1302 (!requeue_pi
&& this->rt_waiter
)) {
1308 * Wake nr_wake waiters. For requeue_pi, if we acquired the
1309 * lock, we already woke the top_waiter. If not, it will be
1310 * woken by futex_unlock_pi().
1312 if (++task_count
<= nr_wake
&& !requeue_pi
) {
1317 /* Ensure we requeue to the expected futex for requeue_pi. */
1318 if (requeue_pi
&& !match_futex(this->requeue_pi_key
, &key2
)) {
1324 * Requeue nr_requeue waiters and possibly one more in the case
1325 * of requeue_pi if we couldn't acquire the lock atomically.
1328 /* Prepare the waiter to take the rt_mutex. */
1329 atomic_inc(&pi_state
->refcount
);
1330 this->pi_state
= pi_state
;
1331 ret
= rt_mutex_start_proxy_lock(&pi_state
->pi_mutex
,
1335 /* We got the lock. */
1336 requeue_pi_wake_futex(this, &key2
, hb2
);
1341 this->pi_state
= NULL
;
1342 free_pi_state(pi_state
);
1346 requeue_futex(this, hb1
, hb2
, &key2
);
1351 double_unlock_hb(hb1
, hb2
);
1354 * drop_futex_key_refs() must be called outside the spinlocks. During
1355 * the requeue we moved futex_q's from the hash bucket at key1 to the
1356 * one at key2 and updated their key pointer. We no longer need to
1357 * hold the references to key1.
1359 while (--drop_count
>= 0)
1360 drop_futex_key_refs(&key1
);
1363 put_futex_key(fshared
, &key2
);
1365 put_futex_key(fshared
, &key1
);
1367 if (pi_state
!= NULL
)
1368 free_pi_state(pi_state
);
1369 return ret
? ret
: task_count
;
1372 /* The key must be already stored in q->key. */
1373 static inline struct futex_hash_bucket
*queue_lock(struct futex_q
*q
)
1375 struct futex_hash_bucket
*hb
;
1377 get_futex_key_refs(&q
->key
);
1378 hb
= hash_futex(&q
->key
);
1379 q
->lock_ptr
= &hb
->lock
;
1381 spin_lock(&hb
->lock
);
1385 static inline void queue_me(struct futex_q
*q
, struct futex_hash_bucket
*hb
)
1390 * The priority used to register this element is
1391 * - either the real thread-priority for the real-time threads
1392 * (i.e. threads with a priority lower than MAX_RT_PRIO)
1393 * - or MAX_RT_PRIO for non-RT threads.
1394 * Thus, all RT-threads are woken first in priority order, and
1395 * the others are woken last, in FIFO order.
1397 prio
= min(current
->normal_prio
, MAX_RT_PRIO
);
1399 plist_node_init(&q
->list
, prio
);
1400 #ifdef CONFIG_DEBUG_PI_LIST
1401 q
->list
.plist
.lock
= &hb
->lock
;
1403 plist_add(&q
->list
, &hb
->chain
);
1405 spin_unlock(&hb
->lock
);
1409 queue_unlock(struct futex_q
*q
, struct futex_hash_bucket
*hb
)
1411 spin_unlock(&hb
->lock
);
1412 drop_futex_key_refs(&q
->key
);
1416 * queue_me and unqueue_me must be called as a pair, each
1417 * exactly once. They are called with the hashed spinlock held.
1420 /* Return 1 if we were still queued (ie. 0 means we were woken) */
1421 static int unqueue_me(struct futex_q
*q
)
1423 spinlock_t
*lock_ptr
;
1426 /* In the common case we don't take the spinlock, which is nice. */
1428 lock_ptr
= q
->lock_ptr
;
1430 if (lock_ptr
!= NULL
) {
1431 spin_lock(lock_ptr
);
1433 * q->lock_ptr can change between reading it and
1434 * spin_lock(), causing us to take the wrong lock. This
1435 * corrects the race condition.
1437 * Reasoning goes like this: if we have the wrong lock,
1438 * q->lock_ptr must have changed (maybe several times)
1439 * between reading it and the spin_lock(). It can
1440 * change again after the spin_lock() but only if it was
1441 * already changed before the spin_lock(). It cannot,
1442 * however, change back to the original value. Therefore
1443 * we can detect whether we acquired the correct lock.
1445 if (unlikely(lock_ptr
!= q
->lock_ptr
)) {
1446 spin_unlock(lock_ptr
);
1449 WARN_ON(plist_node_empty(&q
->list
));
1450 plist_del(&q
->list
, &q
->list
.plist
);
1452 BUG_ON(q
->pi_state
);
1454 spin_unlock(lock_ptr
);
1458 drop_futex_key_refs(&q
->key
);
1463 * PI futexes can not be requeued and must remove themself from the
1464 * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry
1467 static void unqueue_me_pi(struct futex_q
*q
)
1469 WARN_ON(plist_node_empty(&q
->list
));
1470 plist_del(&q
->list
, &q
->list
.plist
);
1472 BUG_ON(!q
->pi_state
);
1473 free_pi_state(q
->pi_state
);
1476 spin_unlock(q
->lock_ptr
);
1478 drop_futex_key_refs(&q
->key
);
1482 * Fixup the pi_state owner with the new owner.
1484 * Must be called with hash bucket lock held and mm->sem held for non
1487 static int fixup_pi_state_owner(u32 __user
*uaddr
, struct futex_q
*q
,
1488 struct task_struct
*newowner
, int fshared
)
1490 u32 newtid
= task_pid_vnr(newowner
) | FUTEX_WAITERS
;
1491 struct futex_pi_state
*pi_state
= q
->pi_state
;
1492 struct task_struct
*oldowner
= pi_state
->owner
;
1493 u32 uval
, curval
, newval
;
1497 if (!pi_state
->owner
)
1498 newtid
|= FUTEX_OWNER_DIED
;
1501 * We are here either because we stole the rtmutex from the
1502 * pending owner or we are the pending owner which failed to
1503 * get the rtmutex. We have to replace the pending owner TID
1504 * in the user space variable. This must be atomic as we have
1505 * to preserve the owner died bit here.
1507 * Note: We write the user space value _before_ changing the pi_state
1508 * because we can fault here. Imagine swapped out pages or a fork
1509 * that marked all the anonymous memory readonly for cow.
1511 * Modifying pi_state _before_ the user space value would
1512 * leave the pi_state in an inconsistent state when we fault
1513 * here, because we need to drop the hash bucket lock to
1514 * handle the fault. This might be observed in the PID check
1515 * in lookup_pi_state.
1518 if (get_futex_value_locked(&uval
, uaddr
))
1522 newval
= (uval
& FUTEX_OWNER_DIED
) | newtid
;
1524 curval
= cmpxchg_futex_value_locked(uaddr
, uval
, newval
);
1526 if (curval
== -EFAULT
)
1534 * We fixed up user space. Now we need to fix the pi_state
1537 if (pi_state
->owner
!= NULL
) {
1538 spin_lock_irq(&pi_state
->owner
->pi_lock
);
1539 WARN_ON(list_empty(&pi_state
->list
));
1540 list_del_init(&pi_state
->list
);
1541 spin_unlock_irq(&pi_state
->owner
->pi_lock
);
1544 pi_state
->owner
= newowner
;
1546 spin_lock_irq(&newowner
->pi_lock
);
1547 WARN_ON(!list_empty(&pi_state
->list
));
1548 list_add(&pi_state
->list
, &newowner
->pi_state_list
);
1549 spin_unlock_irq(&newowner
->pi_lock
);
1553 * To handle the page fault we need to drop the hash bucket
1554 * lock here. That gives the other task (either the pending
1555 * owner itself or the task which stole the rtmutex) the
1556 * chance to try the fixup of the pi_state. So once we are
1557 * back from handling the fault we need to check the pi_state
1558 * after reacquiring the hash bucket lock and before trying to
1559 * do another fixup. When the fixup has been done already we
1563 spin_unlock(q
->lock_ptr
);
1565 ret
= fault_in_user_writeable(uaddr
);
1567 spin_lock(q
->lock_ptr
);
1570 * Check if someone else fixed it for us:
1572 if (pi_state
->owner
!= oldowner
)
1582 * In case we must use restart_block to restart a futex_wait,
1583 * we encode in the 'flags' shared capability
1585 #define FLAGS_SHARED 0x01
1586 #define FLAGS_CLOCKRT 0x02
1587 #define FLAGS_HAS_TIMEOUT 0x04
1589 static long futex_wait_restart(struct restart_block
*restart
);
1592 * fixup_owner() - Post lock pi_state and corner case management
1593 * @uaddr: user address of the futex
1594 * @fshared: whether the futex is shared (1) or not (0)
1595 * @q: futex_q (contains pi_state and access to the rt_mutex)
1596 * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0)
1598 * After attempting to lock an rt_mutex, this function is called to cleanup
1599 * the pi_state owner as well as handle race conditions that may allow us to
1600 * acquire the lock. Must be called with the hb lock held.
1603 * 1 - success, lock taken
1604 * 0 - success, lock not taken
1605 * <0 - on error (-EFAULT)
1607 static int fixup_owner(u32 __user
*uaddr
, int fshared
, struct futex_q
*q
,
1610 struct task_struct
*owner
;
1615 * Got the lock. We might not be the anticipated owner if we
1616 * did a lock-steal - fix up the PI-state in that case:
1618 if (q
->pi_state
->owner
!= current
)
1619 ret
= fixup_pi_state_owner(uaddr
, q
, current
, fshared
);
1624 * Catch the rare case, where the lock was released when we were on the
1625 * way back before we locked the hash bucket.
1627 if (q
->pi_state
->owner
== current
) {
1629 * Try to get the rt_mutex now. This might fail as some other
1630 * task acquired the rt_mutex after we removed ourself from the
1631 * rt_mutex waiters list.
1633 if (rt_mutex_trylock(&q
->pi_state
->pi_mutex
)) {
1639 * pi_state is incorrect, some other task did a lock steal and
1640 * we returned due to timeout or signal without taking the
1641 * rt_mutex. Too late. We can access the rt_mutex_owner without
1642 * locking, as the other task is now blocked on the hash bucket
1643 * lock. Fix the state up.
1645 owner
= rt_mutex_owner(&q
->pi_state
->pi_mutex
);
1646 ret
= fixup_pi_state_owner(uaddr
, q
, owner
, fshared
);
1651 * Paranoia check. If we did not take the lock, then we should not be
1652 * the owner, nor the pending owner, of the rt_mutex.
1654 if (rt_mutex_owner(&q
->pi_state
->pi_mutex
) == current
)
1655 printk(KERN_ERR
"fixup_owner: ret = %d pi-mutex: %p "
1656 "pi-state %p\n", ret
,
1657 q
->pi_state
->pi_mutex
.owner
,
1658 q
->pi_state
->owner
);
1661 return ret
? ret
: locked
;
1665 * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal
1666 * @hb: the futex hash bucket, must be locked by the caller
1667 * @q: the futex_q to queue up on
1668 * @timeout: the prepared hrtimer_sleeper, or null for no timeout
1670 static void futex_wait_queue_me(struct futex_hash_bucket
*hb
, struct futex_q
*q
,
1671 struct hrtimer_sleeper
*timeout
)
1673 set_current_state(TASK_INTERRUPTIBLE
);
1678 hrtimer_start_expires(&timeout
->timer
, HRTIMER_MODE_ABS
);
1679 if (!hrtimer_active(&timeout
->timer
))
1680 timeout
->task
= NULL
;
1684 * If we have been removed from the hash list, then another task
1685 * has tried to wake us, and we can skip the call to schedule().
1687 if (likely(!plist_node_empty(&q
->list
))) {
1689 * If the timer has already expired, current will already be
1690 * flagged for rescheduling. Only call schedule if there
1691 * is no timeout, or if it has yet to expire.
1693 if (!timeout
|| timeout
->task
)
1696 __set_current_state(TASK_RUNNING
);
1700 * futex_wait_setup() - Prepare to wait on a futex
1701 * @uaddr: the futex userspace address
1702 * @val: the expected value
1703 * @fshared: whether the futex is shared (1) or not (0)
1704 * @q: the associated futex_q
1705 * @hb: storage for hash_bucket pointer to be returned to caller
1707 * Setup the futex_q and locate the hash_bucket. Get the futex value and
1708 * compare it with the expected value. Handle atomic faults internally.
1709 * Return with the hb lock held and a q.key reference on success, and unlocked
1710 * with no q.key reference on failure.
1713 * 0 - uaddr contains val and hb has been locked
1714 * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlcoked
1716 static int futex_wait_setup(u32 __user
*uaddr
, u32 val
, int fshared
,
1717 struct futex_q
*q
, struct futex_hash_bucket
**hb
)
1723 * Access the page AFTER the hash-bucket is locked.
1724 * Order is important:
1726 * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val);
1727 * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); }
1729 * The basic logical guarantee of a futex is that it blocks ONLY
1730 * if cond(var) is known to be true at the time of blocking, for
1731 * any cond. If we queued after testing *uaddr, that would open
1732 * a race condition where we could block indefinitely with
1733 * cond(var) false, which would violate the guarantee.
1735 * A consequence is that futex_wait() can return zero and absorb
1736 * a wakeup when *uaddr != val on entry to the syscall. This is
1740 q
->key
= FUTEX_KEY_INIT
;
1741 ret
= get_futex_key(uaddr
, fshared
, &q
->key
, VERIFY_READ
);
1742 if (unlikely(ret
!= 0))
1746 *hb
= queue_lock(q
);
1748 ret
= get_futex_value_locked(&uval
, uaddr
);
1751 queue_unlock(q
, *hb
);
1753 ret
= get_user(uval
, uaddr
);
1760 put_futex_key(fshared
, &q
->key
);
1765 queue_unlock(q
, *hb
);
1771 put_futex_key(fshared
, &q
->key
);
1775 static int futex_wait(u32 __user
*uaddr
, int fshared
,
1776 u32 val
, ktime_t
*abs_time
, u32 bitset
, int clockrt
)
1778 struct hrtimer_sleeper timeout
, *to
= NULL
;
1779 struct restart_block
*restart
;
1780 struct futex_hash_bucket
*hb
;
1790 q
.requeue_pi_key
= NULL
;
1795 hrtimer_init_on_stack(&to
->timer
, clockrt
? CLOCK_REALTIME
:
1796 CLOCK_MONOTONIC
, HRTIMER_MODE_ABS
);
1797 hrtimer_init_sleeper(to
, current
);
1798 hrtimer_set_expires_range_ns(&to
->timer
, *abs_time
,
1799 current
->timer_slack_ns
);
1803 /* Prepare to wait on uaddr. */
1804 ret
= futex_wait_setup(uaddr
, val
, fshared
, &q
, &hb
);
1808 /* queue_me and wait for wakeup, timeout, or a signal. */
1809 futex_wait_queue_me(hb
, &q
, to
);
1811 /* If we were woken (and unqueued), we succeeded, whatever. */
1813 if (!unqueue_me(&q
))
1816 if (to
&& !to
->task
)
1820 * We expect signal_pending(current), but we might be the
1821 * victim of a spurious wakeup as well.
1823 if (!signal_pending(current
)) {
1824 put_futex_key(fshared
, &q
.key
);
1832 restart
= ¤t_thread_info()->restart_block
;
1833 restart
->fn
= futex_wait_restart
;
1834 restart
->futex
.uaddr
= (u32
*)uaddr
;
1835 restart
->futex
.val
= val
;
1836 restart
->futex
.time
= abs_time
->tv64
;
1837 restart
->futex
.bitset
= bitset
;
1838 restart
->futex
.flags
= FLAGS_HAS_TIMEOUT
;
1841 restart
->futex
.flags
|= FLAGS_SHARED
;
1843 restart
->futex
.flags
|= FLAGS_CLOCKRT
;
1845 ret
= -ERESTART_RESTARTBLOCK
;
1848 put_futex_key(fshared
, &q
.key
);
1851 hrtimer_cancel(&to
->timer
);
1852 destroy_hrtimer_on_stack(&to
->timer
);
1858 static long futex_wait_restart(struct restart_block
*restart
)
1860 u32 __user
*uaddr
= (u32 __user
*)restart
->futex
.uaddr
;
1862 ktime_t t
, *tp
= NULL
;
1864 if (restart
->futex
.flags
& FLAGS_HAS_TIMEOUT
) {
1865 t
.tv64
= restart
->futex
.time
;
1868 restart
->fn
= do_no_restart_syscall
;
1869 if (restart
->futex
.flags
& FLAGS_SHARED
)
1871 return (long)futex_wait(uaddr
, fshared
, restart
->futex
.val
, tp
,
1872 restart
->futex
.bitset
,
1873 restart
->futex
.flags
& FLAGS_CLOCKRT
);
1878 * Userspace tried a 0 -> TID atomic transition of the futex value
1879 * and failed. The kernel side here does the whole locking operation:
1880 * if there are waiters then it will block, it does PI, etc. (Due to
1881 * races the kernel might see a 0 value of the futex too.)
1883 static int futex_lock_pi(u32 __user
*uaddr
, int fshared
,
1884 int detect
, ktime_t
*time
, int trylock
)
1886 struct hrtimer_sleeper timeout
, *to
= NULL
;
1887 struct futex_hash_bucket
*hb
;
1891 if (refill_pi_state_cache())
1896 hrtimer_init_on_stack(&to
->timer
, CLOCK_REALTIME
,
1898 hrtimer_init_sleeper(to
, current
);
1899 hrtimer_set_expires(&to
->timer
, *time
);
1904 q
.requeue_pi_key
= NULL
;
1906 q
.key
= FUTEX_KEY_INIT
;
1907 ret
= get_futex_key(uaddr
, fshared
, &q
.key
, VERIFY_WRITE
);
1908 if (unlikely(ret
!= 0))
1912 hb
= queue_lock(&q
);
1914 ret
= futex_lock_pi_atomic(uaddr
, hb
, &q
.key
, &q
.pi_state
, current
, 0);
1915 if (unlikely(ret
)) {
1918 /* We got the lock. */
1920 goto out_unlock_put_key
;
1925 * Task is exiting and we just wait for the
1928 queue_unlock(&q
, hb
);
1929 put_futex_key(fshared
, &q
.key
);
1933 goto out_unlock_put_key
;
1938 * Only actually queue now that the atomic ops are done:
1942 WARN_ON(!q
.pi_state
);
1944 * Block on the PI mutex:
1947 ret
= rt_mutex_timed_lock(&q
.pi_state
->pi_mutex
, to
, 1);
1949 ret
= rt_mutex_trylock(&q
.pi_state
->pi_mutex
);
1950 /* Fixup the trylock return value: */
1951 ret
= ret
? 0 : -EWOULDBLOCK
;
1954 spin_lock(q
.lock_ptr
);
1956 * Fixup the pi_state owner and possibly acquire the lock if we
1959 res
= fixup_owner(uaddr
, fshared
, &q
, !ret
);
1961 * If fixup_owner() returned an error, proprogate that. If it acquired
1962 * the lock, clear our -ETIMEDOUT or -EINTR.
1965 ret
= (res
< 0) ? res
: 0;
1968 * If fixup_owner() faulted and was unable to handle the fault, unlock
1969 * it and return the fault to userspace.
1971 if (ret
&& (rt_mutex_owner(&q
.pi_state
->pi_mutex
) == current
))
1972 rt_mutex_unlock(&q
.pi_state
->pi_mutex
);
1974 /* Unqueue and drop the lock */
1980 queue_unlock(&q
, hb
);
1983 put_futex_key(fshared
, &q
.key
);
1986 destroy_hrtimer_on_stack(&to
->timer
);
1987 return ret
!= -EINTR
? ret
: -ERESTARTNOINTR
;
1990 queue_unlock(&q
, hb
);
1992 ret
= fault_in_user_writeable(uaddr
);
1999 put_futex_key(fshared
, &q
.key
);
2004 * Userspace attempted a TID -> 0 atomic transition, and failed.
2005 * This is the in-kernel slowpath: we look up the PI state (if any),
2006 * and do the rt-mutex unlock.
2008 static int futex_unlock_pi(u32 __user
*uaddr
, int fshared
)
2010 struct futex_hash_bucket
*hb
;
2011 struct futex_q
*this, *next
;
2013 struct plist_head
*head
;
2014 union futex_key key
= FUTEX_KEY_INIT
;
2018 if (get_user(uval
, uaddr
))
2021 * We release only a lock we actually own:
2023 if ((uval
& FUTEX_TID_MASK
) != task_pid_vnr(current
))
2026 ret
= get_futex_key(uaddr
, fshared
, &key
, VERIFY_WRITE
);
2027 if (unlikely(ret
!= 0))
2030 hb
= hash_futex(&key
);
2031 spin_lock(&hb
->lock
);
2034 * To avoid races, try to do the TID -> 0 atomic transition
2035 * again. If it succeeds then we can return without waking
2038 if (!(uval
& FUTEX_OWNER_DIED
))
2039 uval
= cmpxchg_futex_value_locked(uaddr
, task_pid_vnr(current
), 0);
2042 if (unlikely(uval
== -EFAULT
))
2045 * Rare case: we managed to release the lock atomically,
2046 * no need to wake anyone else up:
2048 if (unlikely(uval
== task_pid_vnr(current
)))
2052 * Ok, other tasks may need to be woken up - check waiters
2053 * and do the wakeup if necessary:
2057 plist_for_each_entry_safe(this, next
, head
, list
) {
2058 if (!match_futex (&this->key
, &key
))
2060 ret
= wake_futex_pi(uaddr
, uval
, this);
2062 * The atomic access to the futex value
2063 * generated a pagefault, so retry the
2064 * user-access and the wakeup:
2071 * No waiters - kernel unlocks the futex:
2073 if (!(uval
& FUTEX_OWNER_DIED
)) {
2074 ret
= unlock_futex_pi(uaddr
, uval
);
2080 spin_unlock(&hb
->lock
);
2081 put_futex_key(fshared
, &key
);
2087 spin_unlock(&hb
->lock
);
2088 put_futex_key(fshared
, &key
);
2090 ret
= fault_in_user_writeable(uaddr
);
2098 * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex
2099 * @hb: the hash_bucket futex_q was original enqueued on
2100 * @q: the futex_q woken while waiting to be requeued
2101 * @key2: the futex_key of the requeue target futex
2102 * @timeout: the timeout associated with the wait (NULL if none)
2104 * Detect if the task was woken on the initial futex as opposed to the requeue
2105 * target futex. If so, determine if it was a timeout or a signal that caused
2106 * the wakeup and return the appropriate error code to the caller. Must be
2107 * called with the hb lock held.
2110 * 0 - no early wakeup detected
2111 * <0 - -ETIMEDOUT or -ERESTARTNOINTR
2114 int handle_early_requeue_pi_wakeup(struct futex_hash_bucket
*hb
,
2115 struct futex_q
*q
, union futex_key
*key2
,
2116 struct hrtimer_sleeper
*timeout
)
2121 * With the hb lock held, we avoid races while we process the wakeup.
2122 * We only need to hold hb (and not hb2) to ensure atomicity as the
2123 * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb.
2124 * It can't be requeued from uaddr2 to something else since we don't
2125 * support a PI aware source futex for requeue.
2127 if (!match_futex(&q
->key
, key2
)) {
2128 WARN_ON(q
->lock_ptr
&& (&hb
->lock
!= q
->lock_ptr
));
2130 * We were woken prior to requeue by a timeout or a signal.
2131 * Unqueue the futex_q and determine which it was.
2133 plist_del(&q
->list
, &q
->list
.plist
);
2135 /* Handle spurious wakeups gracefully */
2137 if (timeout
&& !timeout
->task
)
2139 else if (signal_pending(current
))
2140 ret
= -ERESTARTNOINTR
;
2146 * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2
2147 * @uaddr: the futex we initialyl wait on (non-pi)
2148 * @fshared: whether the futexes are shared (1) or not (0). They must be
2149 * the same type, no requeueing from private to shared, etc.
2150 * @val: the expected value of uaddr
2151 * @abs_time: absolute timeout
2152 * @bitset: 32 bit wakeup bitset set by userspace, defaults to all.
2153 * @clockrt: whether to use CLOCK_REALTIME (1) or CLOCK_MONOTONIC (0)
2154 * @uaddr2: the pi futex we will take prior to returning to user-space
2156 * The caller will wait on uaddr and will be requeued by futex_requeue() to
2157 * uaddr2 which must be PI aware. Normal wakeup will wake on uaddr2 and
2158 * complete the acquisition of the rt_mutex prior to returning to userspace.
2159 * This ensures the rt_mutex maintains an owner when it has waiters; without
2160 * one, the pi logic wouldn't know which task to boost/deboost, if there was a
2163 * We call schedule in futex_wait_queue_me() when we enqueue and return there
2164 * via the following:
2165 * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue()
2166 * 2) wakeup on uaddr2 after a requeue and subsequent unlock
2167 * 3) signal (before or after requeue)
2168 * 4) timeout (before or after requeue)
2170 * If 3, we setup a restart_block with futex_wait_requeue_pi() as the function.
2172 * If 2, we may then block on trying to take the rt_mutex and return via:
2173 * 5) successful lock
2176 * 8) other lock acquisition failure
2178 * If 6, we setup a restart_block with futex_lock_pi() as the function.
2180 * If 4 or 7, we cleanup and return with -ETIMEDOUT.
2186 static int futex_wait_requeue_pi(u32 __user
*uaddr
, int fshared
,
2187 u32 val
, ktime_t
*abs_time
, u32 bitset
,
2188 int clockrt
, u32 __user
*uaddr2
)
2190 struct hrtimer_sleeper timeout
, *to
= NULL
;
2191 struct rt_mutex_waiter rt_waiter
;
2192 struct rt_mutex
*pi_mutex
= NULL
;
2193 struct futex_hash_bucket
*hb
;
2194 union futex_key key2
;
2203 hrtimer_init_on_stack(&to
->timer
, clockrt
? CLOCK_REALTIME
:
2204 CLOCK_MONOTONIC
, HRTIMER_MODE_ABS
);
2205 hrtimer_init_sleeper(to
, current
);
2206 hrtimer_set_expires_range_ns(&to
->timer
, *abs_time
,
2207 current
->timer_slack_ns
);
2211 * The waiter is allocated on our stack, manipulated by the requeue
2212 * code while we sleep on uaddr.
2214 debug_rt_mutex_init_waiter(&rt_waiter
);
2215 rt_waiter
.task
= NULL
;
2217 key2
= FUTEX_KEY_INIT
;
2218 ret
= get_futex_key(uaddr2
, fshared
, &key2
, VERIFY_WRITE
);
2219 if (unlikely(ret
!= 0))
2224 q
.rt_waiter
= &rt_waiter
;
2225 q
.requeue_pi_key
= &key2
;
2227 /* Prepare to wait on uaddr. */
2228 ret
= futex_wait_setup(uaddr
, val
, fshared
, &q
, &hb
);
2232 /* Queue the futex_q, drop the hb lock, wait for wakeup. */
2233 futex_wait_queue_me(hb
, &q
, to
);
2235 spin_lock(&hb
->lock
);
2236 ret
= handle_early_requeue_pi_wakeup(hb
, &q
, &key2
, to
);
2237 spin_unlock(&hb
->lock
);
2242 * In order for us to be here, we know our q.key == key2, and since
2243 * we took the hb->lock above, we also know that futex_requeue() has
2244 * completed and we no longer have to concern ourselves with a wakeup
2245 * race with the atomic proxy lock acquition by the requeue code.
2248 /* Check if the requeue code acquired the second futex for us. */
2251 * Got the lock. We might not be the anticipated owner if we
2252 * did a lock-steal - fix up the PI-state in that case.
2254 if (q
.pi_state
&& (q
.pi_state
->owner
!= current
)) {
2255 spin_lock(q
.lock_ptr
);
2256 ret
= fixup_pi_state_owner(uaddr2
, &q
, current
,
2258 spin_unlock(q
.lock_ptr
);
2262 * We have been woken up by futex_unlock_pi(), a timeout, or a
2263 * signal. futex_unlock_pi() will not destroy the lock_ptr nor
2266 WARN_ON(!&q
.pi_state
);
2267 pi_mutex
= &q
.pi_state
->pi_mutex
;
2268 ret
= rt_mutex_finish_proxy_lock(pi_mutex
, to
, &rt_waiter
, 1);
2269 debug_rt_mutex_free_waiter(&rt_waiter
);
2271 spin_lock(q
.lock_ptr
);
2273 * Fixup the pi_state owner and possibly acquire the lock if we
2276 res
= fixup_owner(uaddr2
, fshared
, &q
, !ret
);
2278 * If fixup_owner() returned an error, proprogate that. If it
2279 * acquired the lock, clear our -ETIMEDOUT or -EINTR.
2282 ret
= (res
< 0) ? res
: 0;
2284 /* Unqueue and drop the lock. */
2289 * If fixup_pi_state_owner() faulted and was unable to handle the
2290 * fault, unlock the rt_mutex and return the fault to userspace.
2292 if (ret
== -EFAULT
) {
2293 if (rt_mutex_owner(pi_mutex
) == current
)
2294 rt_mutex_unlock(pi_mutex
);
2295 } else if (ret
== -EINTR
) {
2297 * We've already been requeued, but we have no way to
2298 * restart by calling futex_lock_pi() directly. We
2299 * could restart the syscall, but that will look at
2300 * the user space value and return right away. So we
2301 * drop back with EWOULDBLOCK to tell user space that
2302 * "val" has been changed. That's the same what the
2303 * restart of the syscall would do in
2304 * futex_wait_setup().
2310 put_futex_key(fshared
, &q
.key
);
2312 put_futex_key(fshared
, &key2
);
2316 hrtimer_cancel(&to
->timer
);
2317 destroy_hrtimer_on_stack(&to
->timer
);
2323 * Support for robust futexes: the kernel cleans up held futexes at
2326 * Implementation: user-space maintains a per-thread list of locks it
2327 * is holding. Upon do_exit(), the kernel carefully walks this list,
2328 * and marks all locks that are owned by this thread with the
2329 * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is
2330 * always manipulated with the lock held, so the list is private and
2331 * per-thread. Userspace also maintains a per-thread 'list_op_pending'
2332 * field, to allow the kernel to clean up if the thread dies after
2333 * acquiring the lock, but just before it could have added itself to
2334 * the list. There can only be one such pending lock.
2338 * sys_set_robust_list - set the robust-futex list head of a task
2339 * @head: pointer to the list-head
2340 * @len: length of the list-head, as userspace expects
2342 SYSCALL_DEFINE2(set_robust_list
, struct robust_list_head __user
*, head
,
2345 if (!futex_cmpxchg_enabled
)
2348 * The kernel knows only one size for now:
2350 if (unlikely(len
!= sizeof(*head
)))
2353 current
->robust_list
= head
;
2359 * sys_get_robust_list - get the robust-futex list head of a task
2360 * @pid: pid of the process [zero for current task]
2361 * @head_ptr: pointer to a list-head pointer, the kernel fills it in
2362 * @len_ptr: pointer to a length field, the kernel fills in the header size
2364 SYSCALL_DEFINE3(get_robust_list
, int, pid
,
2365 struct robust_list_head __user
* __user
*, head_ptr
,
2366 size_t __user
*, len_ptr
)
2368 struct robust_list_head __user
*head
;
2370 const struct cred
*cred
= current_cred(), *pcred
;
2372 if (!futex_cmpxchg_enabled
)
2376 head
= current
->robust_list
;
2378 struct task_struct
*p
;
2382 p
= find_task_by_vpid(pid
);
2386 pcred
= __task_cred(p
);
2387 if (cred
->euid
!= pcred
->euid
&&
2388 cred
->euid
!= pcred
->uid
&&
2389 !capable(CAP_SYS_PTRACE
))
2391 head
= p
->robust_list
;
2395 if (put_user(sizeof(*head
), len_ptr
))
2397 return put_user(head
, head_ptr
);
2406 * Process a futex-list entry, check whether it's owned by the
2407 * dying task, and do notification if so:
2409 int handle_futex_death(u32 __user
*uaddr
, struct task_struct
*curr
, int pi
)
2411 u32 uval
, nval
, mval
;
2414 if (get_user(uval
, uaddr
))
2417 if ((uval
& FUTEX_TID_MASK
) == task_pid_vnr(curr
)) {
2419 * Ok, this dying thread is truly holding a futex
2420 * of interest. Set the OWNER_DIED bit atomically
2421 * via cmpxchg, and if the value had FUTEX_WAITERS
2422 * set, wake up a waiter (if any). (We have to do a
2423 * futex_wake() even if OWNER_DIED is already set -
2424 * to handle the rare but possible case of recursive
2425 * thread-death.) The rest of the cleanup is done in
2428 mval
= (uval
& FUTEX_WAITERS
) | FUTEX_OWNER_DIED
;
2429 nval
= futex_atomic_cmpxchg_inatomic(uaddr
, uval
, mval
);
2431 if (nval
== -EFAULT
)
2438 * Wake robust non-PI futexes here. The wakeup of
2439 * PI futexes happens in exit_pi_state():
2441 if (!pi
&& (uval
& FUTEX_WAITERS
))
2442 futex_wake(uaddr
, 1, 1, FUTEX_BITSET_MATCH_ANY
);
2448 * Fetch a robust-list pointer. Bit 0 signals PI futexes:
2450 static inline int fetch_robust_entry(struct robust_list __user
**entry
,
2451 struct robust_list __user
* __user
*head
,
2454 unsigned long uentry
;
2456 if (get_user(uentry
, (unsigned long __user
*)head
))
2459 *entry
= (void __user
*)(uentry
& ~1UL);
2466 * Walk curr->robust_list (very carefully, it's a userspace list!)
2467 * and mark any locks found there dead, and notify any waiters.
2469 * We silently return on any sign of list-walking problem.
2471 void exit_robust_list(struct task_struct
*curr
)
2473 struct robust_list_head __user
*head
= curr
->robust_list
;
2474 struct robust_list __user
*entry
, *next_entry
, *pending
;
2475 unsigned int limit
= ROBUST_LIST_LIMIT
, pi
, next_pi
, pip
;
2476 unsigned long futex_offset
;
2479 if (!futex_cmpxchg_enabled
)
2483 * Fetch the list head (which was registered earlier, via
2484 * sys_set_robust_list()):
2486 if (fetch_robust_entry(&entry
, &head
->list
.next
, &pi
))
2489 * Fetch the relative futex offset:
2491 if (get_user(futex_offset
, &head
->futex_offset
))
2494 * Fetch any possibly pending lock-add first, and handle it
2497 if (fetch_robust_entry(&pending
, &head
->list_op_pending
, &pip
))
2500 next_entry
= NULL
; /* avoid warning with gcc */
2501 while (entry
!= &head
->list
) {
2503 * Fetch the next entry in the list before calling
2504 * handle_futex_death:
2506 rc
= fetch_robust_entry(&next_entry
, &entry
->next
, &next_pi
);
2508 * A pending lock might already be on the list, so
2509 * don't process it twice:
2511 if (entry
!= pending
)
2512 if (handle_futex_death((void __user
*)entry
+ futex_offset
,
2520 * Avoid excessively long or circular lists:
2529 handle_futex_death((void __user
*)pending
+ futex_offset
,
2533 long do_futex(u32 __user
*uaddr
, int op
, u32 val
, ktime_t
*timeout
,
2534 u32 __user
*uaddr2
, u32 val2
, u32 val3
)
2536 int clockrt
, ret
= -ENOSYS
;
2537 int cmd
= op
& FUTEX_CMD_MASK
;
2540 if (!(op
& FUTEX_PRIVATE_FLAG
))
2543 clockrt
= op
& FUTEX_CLOCK_REALTIME
;
2544 if (clockrt
&& cmd
!= FUTEX_WAIT_BITSET
&& cmd
!= FUTEX_WAIT_REQUEUE_PI
)
2549 val3
= FUTEX_BITSET_MATCH_ANY
;
2550 case FUTEX_WAIT_BITSET
:
2551 ret
= futex_wait(uaddr
, fshared
, val
, timeout
, val3
, clockrt
);
2554 val3
= FUTEX_BITSET_MATCH_ANY
;
2555 case FUTEX_WAKE_BITSET
:
2556 ret
= futex_wake(uaddr
, fshared
, val
, val3
);
2559 ret
= futex_requeue(uaddr
, fshared
, uaddr2
, val
, val2
, NULL
, 0);
2561 case FUTEX_CMP_REQUEUE
:
2562 ret
= futex_requeue(uaddr
, fshared
, uaddr2
, val
, val2
, &val3
,
2566 ret
= futex_wake_op(uaddr
, fshared
, uaddr2
, val
, val2
, val3
);
2569 if (futex_cmpxchg_enabled
)
2570 ret
= futex_lock_pi(uaddr
, fshared
, val
, timeout
, 0);
2572 case FUTEX_UNLOCK_PI
:
2573 if (futex_cmpxchg_enabled
)
2574 ret
= futex_unlock_pi(uaddr
, fshared
);
2576 case FUTEX_TRYLOCK_PI
:
2577 if (futex_cmpxchg_enabled
)
2578 ret
= futex_lock_pi(uaddr
, fshared
, 0, timeout
, 1);
2580 case FUTEX_WAIT_REQUEUE_PI
:
2581 val3
= FUTEX_BITSET_MATCH_ANY
;
2582 ret
= futex_wait_requeue_pi(uaddr
, fshared
, val
, timeout
, val3
,
2585 case FUTEX_CMP_REQUEUE_PI
:
2586 ret
= futex_requeue(uaddr
, fshared
, uaddr2
, val
, val2
, &val3
,
2596 SYSCALL_DEFINE6(futex
, u32 __user
*, uaddr
, int, op
, u32
, val
,
2597 struct timespec __user
*, utime
, u32 __user
*, uaddr2
,
2601 ktime_t t
, *tp
= NULL
;
2603 int cmd
= op
& FUTEX_CMD_MASK
;
2605 if (utime
&& (cmd
== FUTEX_WAIT
|| cmd
== FUTEX_LOCK_PI
||
2606 cmd
== FUTEX_WAIT_BITSET
||
2607 cmd
== FUTEX_WAIT_REQUEUE_PI
)) {
2608 if (copy_from_user(&ts
, utime
, sizeof(ts
)) != 0)
2610 if (!timespec_valid(&ts
))
2613 t
= timespec_to_ktime(ts
);
2614 if (cmd
== FUTEX_WAIT
)
2615 t
= ktime_add_safe(ktime_get(), t
);
2619 * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*.
2620 * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP.
2622 if (cmd
== FUTEX_REQUEUE
|| cmd
== FUTEX_CMP_REQUEUE
||
2623 cmd
== FUTEX_CMP_REQUEUE_PI
|| cmd
== FUTEX_WAKE_OP
)
2624 val2
= (u32
) (unsigned long) utime
;
2626 return do_futex(uaddr
, op
, val
, tp
, uaddr2
, val2
, val3
);
2629 static int __init
futex_init(void)
2635 * This will fail and we want it. Some arch implementations do
2636 * runtime detection of the futex_atomic_cmpxchg_inatomic()
2637 * functionality. We want to know that before we call in any
2638 * of the complex code paths. Also we want to prevent
2639 * registration of robust lists in that case. NULL is
2640 * guaranteed to fault and we get -EFAULT on functional
2641 * implementation, the non functional ones will return
2644 curval
= cmpxchg_futex_value_locked(NULL
, 0, 0);
2645 if (curval
== -EFAULT
)
2646 futex_cmpxchg_enabled
= 1;
2648 for (i
= 0; i
< ARRAY_SIZE(futex_queues
); i
++) {
2649 plist_head_init(&futex_queues
[i
].chain
, &futex_queues
[i
].lock
);
2650 spin_lock_init(&futex_queues
[i
].lock
);
2655 __initcall(futex_init
);