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 * Futex flags used to encode options to functions and preserve them across
75 #define FLAGS_SHARED 0x01
76 #define FLAGS_CLOCKRT 0x02
77 #define FLAGS_HAS_TIMEOUT 0x04
80 * Priority Inheritance state:
82 struct futex_pi_state
{
84 * list of 'owned' pi_state instances - these have to be
85 * cleaned up in do_exit() if the task exits prematurely:
87 struct list_head list
;
92 struct rt_mutex pi_mutex
;
94 struct task_struct
*owner
;
101 * struct futex_q - The hashed futex queue entry, one per waiting task
102 * @list: priority-sorted list of tasks waiting on this futex
103 * @task: the task waiting on the futex
104 * @lock_ptr: the hash bucket lock
105 * @key: the key the futex is hashed on
106 * @pi_state: optional priority inheritance state
107 * @rt_waiter: rt_waiter storage for use with requeue_pi
108 * @requeue_pi_key: the requeue_pi target futex key
109 * @bitset: bitset for the optional bitmasked wakeup
111 * We use this hashed waitqueue, instead of a normal wait_queue_t, so
112 * we can wake only the relevant ones (hashed queues may be shared).
114 * A futex_q has a woken state, just like tasks have TASK_RUNNING.
115 * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0.
116 * The order of wakeup is always to make the first condition true, then
119 * PI futexes are typically woken before they are removed from the hash list via
120 * the rt_mutex code. See unqueue_me_pi().
123 struct plist_node list
;
125 struct task_struct
*task
;
126 spinlock_t
*lock_ptr
;
128 struct futex_pi_state
*pi_state
;
129 struct rt_mutex_waiter
*rt_waiter
;
130 union futex_key
*requeue_pi_key
;
134 static const struct futex_q futex_q_init
= {
135 /* list gets initialized in queue_me()*/
136 .key
= FUTEX_KEY_INIT
,
137 .bitset
= FUTEX_BITSET_MATCH_ANY
141 * Hash buckets are shared by all the futex_keys that hash to the same
142 * location. Each key may have multiple futex_q structures, one for each task
143 * waiting on a futex.
145 struct futex_hash_bucket
{
147 struct plist_head chain
;
150 static struct futex_hash_bucket futex_queues
[1<<FUTEX_HASHBITS
];
153 * We hash on the keys returned from get_futex_key (see below).
155 static struct futex_hash_bucket
*hash_futex(union futex_key
*key
)
157 u32 hash
= jhash2((u32
*)&key
->both
.word
,
158 (sizeof(key
->both
.word
)+sizeof(key
->both
.ptr
))/4,
160 return &futex_queues
[hash
& ((1 << FUTEX_HASHBITS
)-1)];
164 * Return 1 if two futex_keys are equal, 0 otherwise.
166 static inline int match_futex(union futex_key
*key1
, union futex_key
*key2
)
169 && key1
->both
.word
== key2
->both
.word
170 && key1
->both
.ptr
== key2
->both
.ptr
171 && key1
->both
.offset
== key2
->both
.offset
);
175 * Take a reference to the resource addressed by a key.
176 * Can be called while holding spinlocks.
179 static void get_futex_key_refs(union futex_key
*key
)
184 switch (key
->both
.offset
& (FUT_OFF_INODE
|FUT_OFF_MMSHARED
)) {
186 ihold(key
->shared
.inode
);
188 case FUT_OFF_MMSHARED
:
189 atomic_inc(&key
->private.mm
->mm_count
);
195 * Drop a reference to the resource addressed by a key.
196 * The hash bucket spinlock must not be held.
198 static void drop_futex_key_refs(union futex_key
*key
)
200 if (!key
->both
.ptr
) {
201 /* If we're here then we tried to put a key we failed to get */
206 switch (key
->both
.offset
& (FUT_OFF_INODE
|FUT_OFF_MMSHARED
)) {
208 iput(key
->shared
.inode
);
210 case FUT_OFF_MMSHARED
:
211 mmdrop(key
->private.mm
);
217 * get_futex_key() - Get parameters which are the keys for a futex
218 * @uaddr: virtual address of the futex
219 * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED
220 * @key: address where result is stored.
222 * Returns a negative error code or 0
223 * The key words are stored in *key on success.
225 * For shared mappings, it's (page->index, vma->vm_file->f_path.dentry->d_inode,
226 * offset_within_page). For private mappings, it's (uaddr, current->mm).
227 * We can usually work out the index without swapping in the page.
229 * lock_page() might sleep, the caller should not hold a spinlock.
232 get_futex_key(u32 __user
*uaddr
, int fshared
, union futex_key
*key
)
234 unsigned long address
= (unsigned long)uaddr
;
235 struct mm_struct
*mm
= current
->mm
;
236 struct page
*page
, *page_head
;
240 * The futex address must be "naturally" aligned.
242 key
->both
.offset
= address
% PAGE_SIZE
;
243 if (unlikely((address
% sizeof(u32
)) != 0))
245 address
-= key
->both
.offset
;
248 * PROCESS_PRIVATE futexes are fast.
249 * As the mm cannot disappear under us and the 'key' only needs
250 * virtual address, we dont even have to find the underlying vma.
251 * Note : We do have to check 'uaddr' is a valid user address,
252 * but access_ok() should be faster than find_vma()
255 if (unlikely(!access_ok(VERIFY_WRITE
, uaddr
, sizeof(u32
))))
257 key
->private.mm
= mm
;
258 key
->private.address
= address
;
259 get_futex_key_refs(key
);
264 err
= get_user_pages_fast(address
, 1, 1, &page
);
268 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
270 if (unlikely(PageTail(page
))) {
272 /* serialize against __split_huge_page_splitting() */
274 if (likely(__get_user_pages_fast(address
, 1, 1, &page
) == 1)) {
275 page_head
= compound_head(page
);
277 * page_head is valid pointer but we must pin
278 * it before taking the PG_lock and/or
279 * PG_compound_lock. The moment we re-enable
280 * irqs __split_huge_page_splitting() can
281 * return and the head page can be freed from
282 * under us. We can't take the PG_lock and/or
283 * PG_compound_lock on a page that could be
284 * freed from under us.
286 if (page
!= page_head
) {
297 page_head
= compound_head(page
);
298 if (page
!= page_head
) {
304 lock_page(page_head
);
305 if (!page_head
->mapping
) {
306 unlock_page(page_head
);
312 * Private mappings are handled in a simple way.
314 * NOTE: When userspace waits on a MAP_SHARED mapping, even if
315 * it's a read-only handle, it's expected that futexes attach to
316 * the object not the particular process.
318 if (PageAnon(page_head
)) {
319 key
->both
.offset
|= FUT_OFF_MMSHARED
; /* ref taken on mm */
320 key
->private.mm
= mm
;
321 key
->private.address
= address
;
323 key
->both
.offset
|= FUT_OFF_INODE
; /* inode-based key */
324 key
->shared
.inode
= page_head
->mapping
->host
;
325 key
->shared
.pgoff
= page_head
->index
;
328 get_futex_key_refs(key
);
330 unlock_page(page_head
);
335 static inline void put_futex_key(union futex_key
*key
)
337 drop_futex_key_refs(key
);
341 * fault_in_user_writeable() - Fault in user address and verify RW access
342 * @uaddr: pointer to faulting user space address
344 * Slow path to fixup the fault we just took in the atomic write
347 * We have no generic implementation of a non-destructive write to the
348 * user address. We know that we faulted in the atomic pagefault
349 * disabled section so we can as well avoid the #PF overhead by
350 * calling get_user_pages() right away.
352 static int fault_in_user_writeable(u32 __user
*uaddr
)
354 struct mm_struct
*mm
= current
->mm
;
357 down_read(&mm
->mmap_sem
);
358 ret
= get_user_pages(current
, mm
, (unsigned long)uaddr
,
359 1, 1, 0, NULL
, NULL
);
360 up_read(&mm
->mmap_sem
);
362 return ret
< 0 ? ret
: 0;
366 * futex_top_waiter() - Return the highest priority waiter on a futex
367 * @hb: the hash bucket the futex_q's reside in
368 * @key: the futex key (to distinguish it from other futex futex_q's)
370 * Must be called with the hb lock held.
372 static struct futex_q
*futex_top_waiter(struct futex_hash_bucket
*hb
,
373 union futex_key
*key
)
375 struct futex_q
*this;
377 plist_for_each_entry(this, &hb
->chain
, list
) {
378 if (match_futex(&this->key
, key
))
384 static u32
cmpxchg_futex_value_locked(u32 __user
*uaddr
, u32 uval
, u32 newval
)
389 curval
= futex_atomic_cmpxchg_inatomic(uaddr
, uval
, newval
);
395 static int get_futex_value_locked(u32
*dest
, u32 __user
*from
)
400 ret
= __copy_from_user_inatomic(dest
, from
, sizeof(u32
));
403 return ret
? -EFAULT
: 0;
410 static int refill_pi_state_cache(void)
412 struct futex_pi_state
*pi_state
;
414 if (likely(current
->pi_state_cache
))
417 pi_state
= kzalloc(sizeof(*pi_state
), GFP_KERNEL
);
422 INIT_LIST_HEAD(&pi_state
->list
);
423 /* pi_mutex gets initialized later */
424 pi_state
->owner
= NULL
;
425 atomic_set(&pi_state
->refcount
, 1);
426 pi_state
->key
= FUTEX_KEY_INIT
;
428 current
->pi_state_cache
= pi_state
;
433 static struct futex_pi_state
* alloc_pi_state(void)
435 struct futex_pi_state
*pi_state
= current
->pi_state_cache
;
438 current
->pi_state_cache
= NULL
;
443 static void free_pi_state(struct futex_pi_state
*pi_state
)
445 if (!atomic_dec_and_test(&pi_state
->refcount
))
449 * If pi_state->owner is NULL, the owner is most probably dying
450 * and has cleaned up the pi_state already
452 if (pi_state
->owner
) {
453 raw_spin_lock_irq(&pi_state
->owner
->pi_lock
);
454 list_del_init(&pi_state
->list
);
455 raw_spin_unlock_irq(&pi_state
->owner
->pi_lock
);
457 rt_mutex_proxy_unlock(&pi_state
->pi_mutex
, pi_state
->owner
);
460 if (current
->pi_state_cache
)
464 * pi_state->list is already empty.
465 * clear pi_state->owner.
466 * refcount is at 0 - put it back to 1.
468 pi_state
->owner
= NULL
;
469 atomic_set(&pi_state
->refcount
, 1);
470 current
->pi_state_cache
= pi_state
;
475 * Look up the task based on what TID userspace gave us.
478 static struct task_struct
* futex_find_get_task(pid_t pid
)
480 struct task_struct
*p
;
483 p
= find_task_by_vpid(pid
);
493 * This task is holding PI mutexes at exit time => bad.
494 * Kernel cleans up PI-state, but userspace is likely hosed.
495 * (Robust-futex cleanup is separate and might save the day for userspace.)
497 void exit_pi_state_list(struct task_struct
*curr
)
499 struct list_head
*next
, *head
= &curr
->pi_state_list
;
500 struct futex_pi_state
*pi_state
;
501 struct futex_hash_bucket
*hb
;
502 union futex_key key
= FUTEX_KEY_INIT
;
504 if (!futex_cmpxchg_enabled
)
507 * We are a ZOMBIE and nobody can enqueue itself on
508 * pi_state_list anymore, but we have to be careful
509 * versus waiters unqueueing themselves:
511 raw_spin_lock_irq(&curr
->pi_lock
);
512 while (!list_empty(head
)) {
515 pi_state
= list_entry(next
, struct futex_pi_state
, list
);
517 hb
= hash_futex(&key
);
518 raw_spin_unlock_irq(&curr
->pi_lock
);
520 spin_lock(&hb
->lock
);
522 raw_spin_lock_irq(&curr
->pi_lock
);
524 * We dropped the pi-lock, so re-check whether this
525 * task still owns the PI-state:
527 if (head
->next
!= next
) {
528 spin_unlock(&hb
->lock
);
532 WARN_ON(pi_state
->owner
!= curr
);
533 WARN_ON(list_empty(&pi_state
->list
));
534 list_del_init(&pi_state
->list
);
535 pi_state
->owner
= NULL
;
536 raw_spin_unlock_irq(&curr
->pi_lock
);
538 rt_mutex_unlock(&pi_state
->pi_mutex
);
540 spin_unlock(&hb
->lock
);
542 raw_spin_lock_irq(&curr
->pi_lock
);
544 raw_spin_unlock_irq(&curr
->pi_lock
);
548 lookup_pi_state(u32 uval
, struct futex_hash_bucket
*hb
,
549 union futex_key
*key
, struct futex_pi_state
**ps
)
551 struct futex_pi_state
*pi_state
= NULL
;
552 struct futex_q
*this, *next
;
553 struct plist_head
*head
;
554 struct task_struct
*p
;
555 pid_t pid
= uval
& FUTEX_TID_MASK
;
559 plist_for_each_entry_safe(this, next
, head
, list
) {
560 if (match_futex(&this->key
, key
)) {
562 * Another waiter already exists - bump up
563 * the refcount and return its pi_state:
565 pi_state
= this->pi_state
;
567 * Userspace might have messed up non-PI and PI futexes
569 if (unlikely(!pi_state
))
572 WARN_ON(!atomic_read(&pi_state
->refcount
));
575 * When pi_state->owner is NULL then the owner died
576 * and another waiter is on the fly. pi_state->owner
577 * is fixed up by the task which acquires
578 * pi_state->rt_mutex.
580 * We do not check for pid == 0 which can happen when
581 * the owner died and robust_list_exit() cleared the
584 if (pid
&& pi_state
->owner
) {
586 * Bail out if user space manipulated the
589 if (pid
!= task_pid_vnr(pi_state
->owner
))
593 atomic_inc(&pi_state
->refcount
);
601 * We are the first waiter - try to look up the real owner and attach
602 * the new pi_state to it, but bail out when TID = 0
606 p
= futex_find_get_task(pid
);
611 * We need to look at the task state flags to figure out,
612 * whether the task is exiting. To protect against the do_exit
613 * change of the task flags, we do this protected by
616 raw_spin_lock_irq(&p
->pi_lock
);
617 if (unlikely(p
->flags
& PF_EXITING
)) {
619 * The task is on the way out. When PF_EXITPIDONE is
620 * set, we know that the task has finished the
623 int ret
= (p
->flags
& PF_EXITPIDONE
) ? -ESRCH
: -EAGAIN
;
625 raw_spin_unlock_irq(&p
->pi_lock
);
630 pi_state
= alloc_pi_state();
633 * Initialize the pi_mutex in locked state and make 'p'
636 rt_mutex_init_proxy_locked(&pi_state
->pi_mutex
, p
);
638 /* Store the key for possible exit cleanups: */
639 pi_state
->key
= *key
;
641 WARN_ON(!list_empty(&pi_state
->list
));
642 list_add(&pi_state
->list
, &p
->pi_state_list
);
644 raw_spin_unlock_irq(&p
->pi_lock
);
654 * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex
655 * @uaddr: the pi futex user address
656 * @hb: the pi futex hash bucket
657 * @key: the futex key associated with uaddr and hb
658 * @ps: the pi_state pointer where we store the result of the
660 * @task: the task to perform the atomic lock work for. This will
661 * be "current" except in the case of requeue pi.
662 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
666 * 1 - acquired the lock
669 * The hb->lock and futex_key refs shall be held by the caller.
671 static int futex_lock_pi_atomic(u32 __user
*uaddr
, struct futex_hash_bucket
*hb
,
672 union futex_key
*key
,
673 struct futex_pi_state
**ps
,
674 struct task_struct
*task
, int set_waiters
)
676 int lock_taken
, ret
, ownerdied
= 0;
677 u32 uval
, newval
, curval
;
680 ret
= lock_taken
= 0;
683 * To avoid races, we attempt to take the lock here again
684 * (by doing a 0 -> TID atomic cmpxchg), while holding all
685 * the locks. It will most likely not succeed.
687 newval
= task_pid_vnr(task
);
689 newval
|= FUTEX_WAITERS
;
691 curval
= cmpxchg_futex_value_locked(uaddr
, 0, newval
);
693 if (unlikely(curval
== -EFAULT
))
699 if ((unlikely((curval
& FUTEX_TID_MASK
) == task_pid_vnr(task
))))
703 * Surprise - we got the lock. Just return to userspace:
705 if (unlikely(!curval
))
711 * Set the FUTEX_WAITERS flag, so the owner will know it has someone
712 * to wake at the next unlock.
714 newval
= curval
| FUTEX_WAITERS
;
717 * There are two cases, where a futex might have no owner (the
718 * owner TID is 0): OWNER_DIED. We take over the futex in this
719 * case. We also do an unconditional take over, when the owner
722 * This is safe as we are protected by the hash bucket lock !
724 if (unlikely(ownerdied
|| !(curval
& FUTEX_TID_MASK
))) {
725 /* Keep the OWNER_DIED bit */
726 newval
= (curval
& ~FUTEX_TID_MASK
) | task_pid_vnr(task
);
731 curval
= cmpxchg_futex_value_locked(uaddr
, uval
, newval
);
733 if (unlikely(curval
== -EFAULT
))
735 if (unlikely(curval
!= uval
))
739 * We took the lock due to owner died take over.
741 if (unlikely(lock_taken
))
745 * We dont have the lock. Look up the PI state (or create it if
746 * we are the first waiter):
748 ret
= lookup_pi_state(uval
, hb
, key
, ps
);
754 * No owner found for this futex. Check if the
755 * OWNER_DIED bit is set to figure out whether
756 * this is a robust futex or not.
758 if (get_futex_value_locked(&curval
, uaddr
))
762 * We simply start over in case of a robust
763 * futex. The code above will take the futex
766 if (curval
& FUTEX_OWNER_DIED
) {
779 * The hash bucket lock must be held when this is called.
780 * Afterwards, the futex_q must not be accessed.
782 static void wake_futex(struct futex_q
*q
)
784 struct task_struct
*p
= q
->task
;
787 * We set q->lock_ptr = NULL _before_ we wake up the task. If
788 * a non-futex wake up happens on another CPU then the task
789 * might exit and p would dereference a non-existing task
790 * struct. Prevent this by holding a reference on p across the
795 plist_del(&q
->list
, &q
->list
.plist
);
797 * The waiting task can free the futex_q as soon as
798 * q->lock_ptr = NULL is written, without taking any locks. A
799 * memory barrier is required here to prevent the following
800 * store to lock_ptr from getting ahead of the plist_del.
805 wake_up_state(p
, TASK_NORMAL
);
809 static int wake_futex_pi(u32 __user
*uaddr
, u32 uval
, struct futex_q
*this)
811 struct task_struct
*new_owner
;
812 struct futex_pi_state
*pi_state
= this->pi_state
;
819 * If current does not own the pi_state then the futex is
820 * inconsistent and user space fiddled with the futex value.
822 if (pi_state
->owner
!= current
)
825 raw_spin_lock(&pi_state
->pi_mutex
.wait_lock
);
826 new_owner
= rt_mutex_next_owner(&pi_state
->pi_mutex
);
829 * It is possible that the next waiter (the one that brought
830 * this owner to the kernel) timed out and is no longer
831 * waiting on the lock.
834 new_owner
= this->task
;
837 * We pass it to the next owner. (The WAITERS bit is always
838 * kept enabled while there is PI state around. We must also
839 * preserve the owner died bit.)
841 if (!(uval
& FUTEX_OWNER_DIED
)) {
844 newval
= FUTEX_WAITERS
| task_pid_vnr(new_owner
);
846 curval
= cmpxchg_futex_value_locked(uaddr
, uval
, newval
);
848 if (curval
== -EFAULT
)
850 else if (curval
!= uval
)
853 raw_spin_unlock(&pi_state
->pi_mutex
.wait_lock
);
858 raw_spin_lock_irq(&pi_state
->owner
->pi_lock
);
859 WARN_ON(list_empty(&pi_state
->list
));
860 list_del_init(&pi_state
->list
);
861 raw_spin_unlock_irq(&pi_state
->owner
->pi_lock
);
863 raw_spin_lock_irq(&new_owner
->pi_lock
);
864 WARN_ON(!list_empty(&pi_state
->list
));
865 list_add(&pi_state
->list
, &new_owner
->pi_state_list
);
866 pi_state
->owner
= new_owner
;
867 raw_spin_unlock_irq(&new_owner
->pi_lock
);
869 raw_spin_unlock(&pi_state
->pi_mutex
.wait_lock
);
870 rt_mutex_unlock(&pi_state
->pi_mutex
);
875 static int unlock_futex_pi(u32 __user
*uaddr
, u32 uval
)
880 * There is no waiter, so we unlock the futex. The owner died
881 * bit has not to be preserved here. We are the owner:
883 oldval
= cmpxchg_futex_value_locked(uaddr
, uval
, 0);
885 if (oldval
== -EFAULT
)
894 * Express the locking dependencies for lockdep:
897 double_lock_hb(struct futex_hash_bucket
*hb1
, struct futex_hash_bucket
*hb2
)
900 spin_lock(&hb1
->lock
);
902 spin_lock_nested(&hb2
->lock
, SINGLE_DEPTH_NESTING
);
903 } else { /* hb1 > hb2 */
904 spin_lock(&hb2
->lock
);
905 spin_lock_nested(&hb1
->lock
, SINGLE_DEPTH_NESTING
);
910 double_unlock_hb(struct futex_hash_bucket
*hb1
, struct futex_hash_bucket
*hb2
)
912 spin_unlock(&hb1
->lock
);
914 spin_unlock(&hb2
->lock
);
918 * Wake up waiters matching bitset queued on this futex (uaddr).
921 futex_wake(u32 __user
*uaddr
, unsigned int flags
, int nr_wake
, u32 bitset
)
923 struct futex_hash_bucket
*hb
;
924 struct futex_q
*this, *next
;
925 struct plist_head
*head
;
926 union futex_key key
= FUTEX_KEY_INIT
;
932 ret
= get_futex_key(uaddr
, flags
& FLAGS_SHARED
, &key
);
933 if (unlikely(ret
!= 0))
936 hb
= hash_futex(&key
);
937 spin_lock(&hb
->lock
);
940 plist_for_each_entry_safe(this, next
, head
, list
) {
941 if (match_futex (&this->key
, &key
)) {
942 if (this->pi_state
|| this->rt_waiter
) {
947 /* Check if one of the bits is set in both bitsets */
948 if (!(this->bitset
& bitset
))
952 if (++ret
>= nr_wake
)
957 spin_unlock(&hb
->lock
);
964 * Wake up all waiters hashed on the physical page that is mapped
965 * to this virtual address:
968 futex_wake_op(u32 __user
*uaddr1
, unsigned int flags
, u32 __user
*uaddr2
,
969 int nr_wake
, int nr_wake2
, int op
)
971 union futex_key key1
= FUTEX_KEY_INIT
, key2
= FUTEX_KEY_INIT
;
972 struct futex_hash_bucket
*hb1
, *hb2
;
973 struct plist_head
*head
;
974 struct futex_q
*this, *next
;
978 ret
= get_futex_key(uaddr1
, flags
& FLAGS_SHARED
, &key1
);
979 if (unlikely(ret
!= 0))
981 ret
= get_futex_key(uaddr2
, flags
& FLAGS_SHARED
, &key2
);
982 if (unlikely(ret
!= 0))
985 hb1
= hash_futex(&key1
);
986 hb2
= hash_futex(&key2
);
989 double_lock_hb(hb1
, hb2
);
990 op_ret
= futex_atomic_op_inuser(op
, uaddr2
);
991 if (unlikely(op_ret
< 0)) {
993 double_unlock_hb(hb1
, hb2
);
997 * we don't get EFAULT from MMU faults if we don't have an MMU,
998 * but we might get them from range checking
1004 if (unlikely(op_ret
!= -EFAULT
)) {
1009 ret
= fault_in_user_writeable(uaddr2
);
1013 if (!(flags
& FLAGS_SHARED
))
1016 put_futex_key(&key2
);
1017 put_futex_key(&key1
);
1023 plist_for_each_entry_safe(this, next
, head
, list
) {
1024 if (match_futex (&this->key
, &key1
)) {
1026 if (++ret
>= nr_wake
)
1035 plist_for_each_entry_safe(this, next
, head
, list
) {
1036 if (match_futex (&this->key
, &key2
)) {
1038 if (++op_ret
>= nr_wake2
)
1045 double_unlock_hb(hb1
, hb2
);
1047 put_futex_key(&key2
);
1049 put_futex_key(&key1
);
1055 * requeue_futex() - Requeue a futex_q from one hb to another
1056 * @q: the futex_q to requeue
1057 * @hb1: the source hash_bucket
1058 * @hb2: the target hash_bucket
1059 * @key2: the new key for the requeued futex_q
1062 void requeue_futex(struct futex_q
*q
, struct futex_hash_bucket
*hb1
,
1063 struct futex_hash_bucket
*hb2
, union futex_key
*key2
)
1067 * If key1 and key2 hash to the same bucket, no need to
1070 if (likely(&hb1
->chain
!= &hb2
->chain
)) {
1071 plist_del(&q
->list
, &hb1
->chain
);
1072 plist_add(&q
->list
, &hb2
->chain
);
1073 q
->lock_ptr
= &hb2
->lock
;
1074 #ifdef CONFIG_DEBUG_PI_LIST
1075 q
->list
.plist
.spinlock
= &hb2
->lock
;
1078 get_futex_key_refs(key2
);
1083 * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue
1085 * @key: the key of the requeue target futex
1086 * @hb: the hash_bucket of the requeue target futex
1088 * During futex_requeue, with requeue_pi=1, it is possible to acquire the
1089 * target futex if it is uncontended or via a lock steal. Set the futex_q key
1090 * to the requeue target futex so the waiter can detect the wakeup on the right
1091 * futex, but remove it from the hb and NULL the rt_waiter so it can detect
1092 * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock
1093 * to protect access to the pi_state to fixup the owner later. Must be called
1094 * with both q->lock_ptr and hb->lock held.
1097 void requeue_pi_wake_futex(struct futex_q
*q
, union futex_key
*key
,
1098 struct futex_hash_bucket
*hb
)
1100 get_futex_key_refs(key
);
1103 WARN_ON(plist_node_empty(&q
->list
));
1104 plist_del(&q
->list
, &q
->list
.plist
);
1106 WARN_ON(!q
->rt_waiter
);
1107 q
->rt_waiter
= NULL
;
1109 q
->lock_ptr
= &hb
->lock
;
1110 #ifdef CONFIG_DEBUG_PI_LIST
1111 q
->list
.plist
.spinlock
= &hb
->lock
;
1114 wake_up_state(q
->task
, TASK_NORMAL
);
1118 * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter
1119 * @pifutex: the user address of the to futex
1120 * @hb1: the from futex hash bucket, must be locked by the caller
1121 * @hb2: the to futex hash bucket, must be locked by the caller
1122 * @key1: the from futex key
1123 * @key2: the to futex key
1124 * @ps: address to store the pi_state pointer
1125 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
1127 * Try and get the lock on behalf of the top waiter if we can do it atomically.
1128 * Wake the top waiter if we succeed. If the caller specified set_waiters,
1129 * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit.
1130 * hb1 and hb2 must be held by the caller.
1133 * 0 - failed to acquire the lock atomicly
1134 * 1 - acquired the lock
1137 static int futex_proxy_trylock_atomic(u32 __user
*pifutex
,
1138 struct futex_hash_bucket
*hb1
,
1139 struct futex_hash_bucket
*hb2
,
1140 union futex_key
*key1
, union futex_key
*key2
,
1141 struct futex_pi_state
**ps
, int set_waiters
)
1143 struct futex_q
*top_waiter
= NULL
;
1147 if (get_futex_value_locked(&curval
, pifutex
))
1151 * Find the top_waiter and determine if there are additional waiters.
1152 * If the caller intends to requeue more than 1 waiter to pifutex,
1153 * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now,
1154 * as we have means to handle the possible fault. If not, don't set
1155 * the bit unecessarily as it will force the subsequent unlock to enter
1158 top_waiter
= futex_top_waiter(hb1
, key1
);
1160 /* There are no waiters, nothing for us to do. */
1164 /* Ensure we requeue to the expected futex. */
1165 if (!match_futex(top_waiter
->requeue_pi_key
, key2
))
1169 * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in
1170 * the contended case or if set_waiters is 1. The pi_state is returned
1171 * in ps in contended cases.
1173 ret
= futex_lock_pi_atomic(pifutex
, hb2
, key2
, ps
, top_waiter
->task
,
1176 requeue_pi_wake_futex(top_waiter
, key2
, hb2
);
1182 * futex_requeue() - Requeue waiters from uaddr1 to uaddr2
1183 * @uaddr1: source futex user address
1184 * @flags: futex flags (FLAGS_SHARED, etc.)
1185 * @uaddr2: target futex user address
1186 * @nr_wake: number of waiters to wake (must be 1 for requeue_pi)
1187 * @nr_requeue: number of waiters to requeue (0-INT_MAX)
1188 * @cmpval: @uaddr1 expected value (or %NULL)
1189 * @requeue_pi: if we are attempting to requeue from a non-pi futex to a
1190 * pi futex (pi to pi requeue is not supported)
1192 * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire
1193 * uaddr2 atomically on behalf of the top waiter.
1196 * >=0 - on success, the number of tasks requeued or woken
1199 static int futex_requeue(u32 __user
*uaddr1
, unsigned int flags
,
1200 u32 __user
*uaddr2
, int nr_wake
, int nr_requeue
,
1201 u32
*cmpval
, int requeue_pi
)
1203 union futex_key key1
= FUTEX_KEY_INIT
, key2
= FUTEX_KEY_INIT
;
1204 int drop_count
= 0, task_count
= 0, ret
;
1205 struct futex_pi_state
*pi_state
= NULL
;
1206 struct futex_hash_bucket
*hb1
, *hb2
;
1207 struct plist_head
*head1
;
1208 struct futex_q
*this, *next
;
1213 * requeue_pi requires a pi_state, try to allocate it now
1214 * without any locks in case it fails.
1216 if (refill_pi_state_cache())
1219 * requeue_pi must wake as many tasks as it can, up to nr_wake
1220 * + nr_requeue, since it acquires the rt_mutex prior to
1221 * returning to userspace, so as to not leave the rt_mutex with
1222 * waiters and no owner. However, second and third wake-ups
1223 * cannot be predicted as they involve race conditions with the
1224 * first wake and a fault while looking up the pi_state. Both
1225 * pthread_cond_signal() and pthread_cond_broadcast() should
1233 if (pi_state
!= NULL
) {
1235 * We will have to lookup the pi_state again, so free this one
1236 * to keep the accounting correct.
1238 free_pi_state(pi_state
);
1242 ret
= get_futex_key(uaddr1
, flags
& FLAGS_SHARED
, &key1
);
1243 if (unlikely(ret
!= 0))
1245 ret
= get_futex_key(uaddr2
, flags
& FLAGS_SHARED
, &key2
);
1246 if (unlikely(ret
!= 0))
1249 hb1
= hash_futex(&key1
);
1250 hb2
= hash_futex(&key2
);
1253 double_lock_hb(hb1
, hb2
);
1255 if (likely(cmpval
!= NULL
)) {
1258 ret
= get_futex_value_locked(&curval
, uaddr1
);
1260 if (unlikely(ret
)) {
1261 double_unlock_hb(hb1
, hb2
);
1263 ret
= get_user(curval
, uaddr1
);
1267 if (!(flags
& FLAGS_SHARED
))
1270 put_futex_key(&key2
);
1271 put_futex_key(&key1
);
1274 if (curval
!= *cmpval
) {
1280 if (requeue_pi
&& (task_count
- nr_wake
< nr_requeue
)) {
1282 * Attempt to acquire uaddr2 and wake the top waiter. If we
1283 * intend to requeue waiters, force setting the FUTEX_WAITERS
1284 * bit. We force this here where we are able to easily handle
1285 * faults rather in the requeue loop below.
1287 ret
= futex_proxy_trylock_atomic(uaddr2
, hb1
, hb2
, &key1
,
1288 &key2
, &pi_state
, nr_requeue
);
1291 * At this point the top_waiter has either taken uaddr2 or is
1292 * waiting on it. If the former, then the pi_state will not
1293 * exist yet, look it up one more time to ensure we have a
1300 ret
= get_futex_value_locked(&curval2
, uaddr2
);
1302 ret
= lookup_pi_state(curval2
, hb2
, &key2
,
1310 double_unlock_hb(hb1
, hb2
);
1311 put_futex_key(&key2
);
1312 put_futex_key(&key1
);
1313 ret
= fault_in_user_writeable(uaddr2
);
1318 /* The owner was exiting, try again. */
1319 double_unlock_hb(hb1
, hb2
);
1320 put_futex_key(&key2
);
1321 put_futex_key(&key1
);
1329 head1
= &hb1
->chain
;
1330 plist_for_each_entry_safe(this, next
, head1
, list
) {
1331 if (task_count
- nr_wake
>= nr_requeue
)
1334 if (!match_futex(&this->key
, &key1
))
1338 * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always
1339 * be paired with each other and no other futex ops.
1341 if ((requeue_pi
&& !this->rt_waiter
) ||
1342 (!requeue_pi
&& this->rt_waiter
)) {
1348 * Wake nr_wake waiters. For requeue_pi, if we acquired the
1349 * lock, we already woke the top_waiter. If not, it will be
1350 * woken by futex_unlock_pi().
1352 if (++task_count
<= nr_wake
&& !requeue_pi
) {
1357 /* Ensure we requeue to the expected futex for requeue_pi. */
1358 if (requeue_pi
&& !match_futex(this->requeue_pi_key
, &key2
)) {
1364 * Requeue nr_requeue waiters and possibly one more in the case
1365 * of requeue_pi if we couldn't acquire the lock atomically.
1368 /* Prepare the waiter to take the rt_mutex. */
1369 atomic_inc(&pi_state
->refcount
);
1370 this->pi_state
= pi_state
;
1371 ret
= rt_mutex_start_proxy_lock(&pi_state
->pi_mutex
,
1375 /* We got the lock. */
1376 requeue_pi_wake_futex(this, &key2
, hb2
);
1381 this->pi_state
= NULL
;
1382 free_pi_state(pi_state
);
1386 requeue_futex(this, hb1
, hb2
, &key2
);
1391 double_unlock_hb(hb1
, hb2
);
1394 * drop_futex_key_refs() must be called outside the spinlocks. During
1395 * the requeue we moved futex_q's from the hash bucket at key1 to the
1396 * one at key2 and updated their key pointer. We no longer need to
1397 * hold the references to key1.
1399 while (--drop_count
>= 0)
1400 drop_futex_key_refs(&key1
);
1403 put_futex_key(&key2
);
1405 put_futex_key(&key1
);
1407 if (pi_state
!= NULL
)
1408 free_pi_state(pi_state
);
1409 return ret
? ret
: task_count
;
1412 /* The key must be already stored in q->key. */
1413 static inline struct futex_hash_bucket
*queue_lock(struct futex_q
*q
)
1414 __acquires(&hb
->lock
)
1416 struct futex_hash_bucket
*hb
;
1418 hb
= hash_futex(&q
->key
);
1419 q
->lock_ptr
= &hb
->lock
;
1421 spin_lock(&hb
->lock
);
1426 queue_unlock(struct futex_q
*q
, struct futex_hash_bucket
*hb
)
1427 __releases(&hb
->lock
)
1429 spin_unlock(&hb
->lock
);
1433 * queue_me() - Enqueue the futex_q on the futex_hash_bucket
1434 * @q: The futex_q to enqueue
1435 * @hb: The destination hash bucket
1437 * The hb->lock must be held by the caller, and is released here. A call to
1438 * queue_me() is typically paired with exactly one call to unqueue_me(). The
1439 * exceptions involve the PI related operations, which may use unqueue_me_pi()
1440 * or nothing if the unqueue is done as part of the wake process and the unqueue
1441 * state is implicit in the state of woken task (see futex_wait_requeue_pi() for
1444 static inline void queue_me(struct futex_q
*q
, struct futex_hash_bucket
*hb
)
1445 __releases(&hb
->lock
)
1450 * The priority used to register this element is
1451 * - either the real thread-priority for the real-time threads
1452 * (i.e. threads with a priority lower than MAX_RT_PRIO)
1453 * - or MAX_RT_PRIO for non-RT threads.
1454 * Thus, all RT-threads are woken first in priority order, and
1455 * the others are woken last, in FIFO order.
1457 prio
= min(current
->normal_prio
, MAX_RT_PRIO
);
1459 plist_node_init(&q
->list
, prio
);
1460 #ifdef CONFIG_DEBUG_PI_LIST
1461 q
->list
.plist
.spinlock
= &hb
->lock
;
1463 plist_add(&q
->list
, &hb
->chain
);
1465 spin_unlock(&hb
->lock
);
1469 * unqueue_me() - Remove the futex_q from its futex_hash_bucket
1470 * @q: The futex_q to unqueue
1472 * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must
1473 * be paired with exactly one earlier call to queue_me().
1476 * 1 - if the futex_q was still queued (and we removed unqueued it)
1477 * 0 - if the futex_q was already removed by the waking thread
1479 static int unqueue_me(struct futex_q
*q
)
1481 spinlock_t
*lock_ptr
;
1484 /* In the common case we don't take the spinlock, which is nice. */
1486 lock_ptr
= q
->lock_ptr
;
1488 if (lock_ptr
!= NULL
) {
1489 spin_lock(lock_ptr
);
1491 * q->lock_ptr can change between reading it and
1492 * spin_lock(), causing us to take the wrong lock. This
1493 * corrects the race condition.
1495 * Reasoning goes like this: if we have the wrong lock,
1496 * q->lock_ptr must have changed (maybe several times)
1497 * between reading it and the spin_lock(). It can
1498 * change again after the spin_lock() but only if it was
1499 * already changed before the spin_lock(). It cannot,
1500 * however, change back to the original value. Therefore
1501 * we can detect whether we acquired the correct lock.
1503 if (unlikely(lock_ptr
!= q
->lock_ptr
)) {
1504 spin_unlock(lock_ptr
);
1507 WARN_ON(plist_node_empty(&q
->list
));
1508 plist_del(&q
->list
, &q
->list
.plist
);
1510 BUG_ON(q
->pi_state
);
1512 spin_unlock(lock_ptr
);
1516 drop_futex_key_refs(&q
->key
);
1521 * PI futexes can not be requeued and must remove themself from the
1522 * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry
1525 static void unqueue_me_pi(struct futex_q
*q
)
1526 __releases(q
->lock_ptr
)
1528 WARN_ON(plist_node_empty(&q
->list
));
1529 plist_del(&q
->list
, &q
->list
.plist
);
1531 BUG_ON(!q
->pi_state
);
1532 free_pi_state(q
->pi_state
);
1535 spin_unlock(q
->lock_ptr
);
1539 * Fixup the pi_state owner with the new owner.
1541 * Must be called with hash bucket lock held and mm->sem held for non
1544 static int fixup_pi_state_owner(u32 __user
*uaddr
, struct futex_q
*q
,
1545 struct task_struct
*newowner
)
1547 u32 newtid
= task_pid_vnr(newowner
) | FUTEX_WAITERS
;
1548 struct futex_pi_state
*pi_state
= q
->pi_state
;
1549 struct task_struct
*oldowner
= pi_state
->owner
;
1550 u32 uval
, curval
, newval
;
1554 if (!pi_state
->owner
)
1555 newtid
|= FUTEX_OWNER_DIED
;
1558 * We are here either because we stole the rtmutex from the
1559 * pending owner or we are the pending owner which failed to
1560 * get the rtmutex. We have to replace the pending owner TID
1561 * in the user space variable. This must be atomic as we have
1562 * to preserve the owner died bit here.
1564 * Note: We write the user space value _before_ changing the pi_state
1565 * because we can fault here. Imagine swapped out pages or a fork
1566 * that marked all the anonymous memory readonly for cow.
1568 * Modifying pi_state _before_ the user space value would
1569 * leave the pi_state in an inconsistent state when we fault
1570 * here, because we need to drop the hash bucket lock to
1571 * handle the fault. This might be observed in the PID check
1572 * in lookup_pi_state.
1575 if (get_futex_value_locked(&uval
, uaddr
))
1579 newval
= (uval
& FUTEX_OWNER_DIED
) | newtid
;
1581 curval
= cmpxchg_futex_value_locked(uaddr
, uval
, newval
);
1583 if (curval
== -EFAULT
)
1591 * We fixed up user space. Now we need to fix the pi_state
1594 if (pi_state
->owner
!= NULL
) {
1595 raw_spin_lock_irq(&pi_state
->owner
->pi_lock
);
1596 WARN_ON(list_empty(&pi_state
->list
));
1597 list_del_init(&pi_state
->list
);
1598 raw_spin_unlock_irq(&pi_state
->owner
->pi_lock
);
1601 pi_state
->owner
= newowner
;
1603 raw_spin_lock_irq(&newowner
->pi_lock
);
1604 WARN_ON(!list_empty(&pi_state
->list
));
1605 list_add(&pi_state
->list
, &newowner
->pi_state_list
);
1606 raw_spin_unlock_irq(&newowner
->pi_lock
);
1610 * To handle the page fault we need to drop the hash bucket
1611 * lock here. That gives the other task (either the pending
1612 * owner itself or the task which stole the rtmutex) the
1613 * chance to try the fixup of the pi_state. So once we are
1614 * back from handling the fault we need to check the pi_state
1615 * after reacquiring the hash bucket lock and before trying to
1616 * do another fixup. When the fixup has been done already we
1620 spin_unlock(q
->lock_ptr
);
1622 ret
= fault_in_user_writeable(uaddr
);
1624 spin_lock(q
->lock_ptr
);
1627 * Check if someone else fixed it for us:
1629 if (pi_state
->owner
!= oldowner
)
1638 static long futex_wait_restart(struct restart_block
*restart
);
1641 * fixup_owner() - Post lock pi_state and corner case management
1642 * @uaddr: user address of the futex
1643 * @q: futex_q (contains pi_state and access to the rt_mutex)
1644 * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0)
1646 * After attempting to lock an rt_mutex, this function is called to cleanup
1647 * the pi_state owner as well as handle race conditions that may allow us to
1648 * acquire the lock. Must be called with the hb lock held.
1651 * 1 - success, lock taken
1652 * 0 - success, lock not taken
1653 * <0 - on error (-EFAULT)
1655 static int fixup_owner(u32 __user
*uaddr
, struct futex_q
*q
, int locked
)
1657 struct task_struct
*owner
;
1662 * Got the lock. We might not be the anticipated owner if we
1663 * did a lock-steal - fix up the PI-state in that case:
1665 if (q
->pi_state
->owner
!= current
)
1666 ret
= fixup_pi_state_owner(uaddr
, q
, current
);
1671 * Catch the rare case, where the lock was released when we were on the
1672 * way back before we locked the hash bucket.
1674 if (q
->pi_state
->owner
== current
) {
1676 * Try to get the rt_mutex now. This might fail as some other
1677 * task acquired the rt_mutex after we removed ourself from the
1678 * rt_mutex waiters list.
1680 if (rt_mutex_trylock(&q
->pi_state
->pi_mutex
)) {
1686 * pi_state is incorrect, some other task did a lock steal and
1687 * we returned due to timeout or signal without taking the
1688 * rt_mutex. Too late. We can access the rt_mutex_owner without
1689 * locking, as the other task is now blocked on the hash bucket
1690 * lock. Fix the state up.
1692 owner
= rt_mutex_owner(&q
->pi_state
->pi_mutex
);
1693 ret
= fixup_pi_state_owner(uaddr
, q
, owner
);
1698 * Paranoia check. If we did not take the lock, then we should not be
1699 * the owner, nor the pending owner, of the rt_mutex.
1701 if (rt_mutex_owner(&q
->pi_state
->pi_mutex
) == current
)
1702 printk(KERN_ERR
"fixup_owner: ret = %d pi-mutex: %p "
1703 "pi-state %p\n", ret
,
1704 q
->pi_state
->pi_mutex
.owner
,
1705 q
->pi_state
->owner
);
1708 return ret
? ret
: locked
;
1712 * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal
1713 * @hb: the futex hash bucket, must be locked by the caller
1714 * @q: the futex_q to queue up on
1715 * @timeout: the prepared hrtimer_sleeper, or null for no timeout
1717 static void futex_wait_queue_me(struct futex_hash_bucket
*hb
, struct futex_q
*q
,
1718 struct hrtimer_sleeper
*timeout
)
1721 * The task state is guaranteed to be set before another task can
1722 * wake it. set_current_state() is implemented using set_mb() and
1723 * queue_me() calls spin_unlock() upon completion, both serializing
1724 * access to the hash list and forcing another memory barrier.
1726 set_current_state(TASK_INTERRUPTIBLE
);
1731 hrtimer_start_expires(&timeout
->timer
, HRTIMER_MODE_ABS
);
1732 if (!hrtimer_active(&timeout
->timer
))
1733 timeout
->task
= NULL
;
1737 * If we have been removed from the hash list, then another task
1738 * has tried to wake us, and we can skip the call to schedule().
1740 if (likely(!plist_node_empty(&q
->list
))) {
1742 * If the timer has already expired, current will already be
1743 * flagged for rescheduling. Only call schedule if there
1744 * is no timeout, or if it has yet to expire.
1746 if (!timeout
|| timeout
->task
)
1749 __set_current_state(TASK_RUNNING
);
1753 * futex_wait_setup() - Prepare to wait on a futex
1754 * @uaddr: the futex userspace address
1755 * @val: the expected value
1756 * @flags: futex flags (FLAGS_SHARED, etc.)
1757 * @q: the associated futex_q
1758 * @hb: storage for hash_bucket pointer to be returned to caller
1760 * Setup the futex_q and locate the hash_bucket. Get the futex value and
1761 * compare it with the expected value. Handle atomic faults internally.
1762 * Return with the hb lock held and a q.key reference on success, and unlocked
1763 * with no q.key reference on failure.
1766 * 0 - uaddr contains val and hb has been locked
1767 * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlcoked
1769 static int futex_wait_setup(u32 __user
*uaddr
, u32 val
, unsigned int flags
,
1770 struct futex_q
*q
, struct futex_hash_bucket
**hb
)
1776 * Access the page AFTER the hash-bucket is locked.
1777 * Order is important:
1779 * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val);
1780 * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); }
1782 * The basic logical guarantee of a futex is that it blocks ONLY
1783 * if cond(var) is known to be true at the time of blocking, for
1784 * any cond. If we locked the hash-bucket after testing *uaddr, that
1785 * would open a race condition where we could block indefinitely with
1786 * cond(var) false, which would violate the guarantee.
1788 * On the other hand, we insert q and release the hash-bucket only
1789 * after testing *uaddr. This guarantees that futex_wait() will NOT
1790 * absorb a wakeup if *uaddr does not match the desired values
1791 * while the syscall executes.
1794 ret
= get_futex_key(uaddr
, flags
& FLAGS_SHARED
, &q
->key
);
1795 if (unlikely(ret
!= 0))
1799 *hb
= queue_lock(q
);
1801 ret
= get_futex_value_locked(&uval
, uaddr
);
1804 queue_unlock(q
, *hb
);
1806 ret
= get_user(uval
, uaddr
);
1810 if (!(flags
& FLAGS_SHARED
))
1813 put_futex_key(&q
->key
);
1818 queue_unlock(q
, *hb
);
1824 put_futex_key(&q
->key
);
1828 static int futex_wait(u32 __user
*uaddr
, unsigned int flags
, u32 val
,
1829 ktime_t
*abs_time
, u32 bitset
)
1831 struct hrtimer_sleeper timeout
, *to
= NULL
;
1832 struct restart_block
*restart
;
1833 struct futex_hash_bucket
*hb
;
1834 struct futex_q q
= futex_q_init
;
1844 hrtimer_init_on_stack(&to
->timer
, (flags
& FLAGS_CLOCKRT
) ?
1845 CLOCK_REALTIME
: CLOCK_MONOTONIC
,
1847 hrtimer_init_sleeper(to
, current
);
1848 hrtimer_set_expires_range_ns(&to
->timer
, *abs_time
,
1849 current
->timer_slack_ns
);
1854 * Prepare to wait on uaddr. On success, holds hb lock and increments
1857 ret
= futex_wait_setup(uaddr
, val
, flags
, &q
, &hb
);
1861 /* queue_me and wait for wakeup, timeout, or a signal. */
1862 futex_wait_queue_me(hb
, &q
, to
);
1864 /* If we were woken (and unqueued), we succeeded, whatever. */
1866 /* unqueue_me() drops q.key ref */
1867 if (!unqueue_me(&q
))
1870 if (to
&& !to
->task
)
1874 * We expect signal_pending(current), but we might be the
1875 * victim of a spurious wakeup as well.
1877 if (!signal_pending(current
))
1884 restart
= ¤t_thread_info()->restart_block
;
1885 restart
->fn
= futex_wait_restart
;
1886 restart
->futex
.uaddr
= uaddr
;
1887 restart
->futex
.val
= val
;
1888 restart
->futex
.time
= abs_time
->tv64
;
1889 restart
->futex
.bitset
= bitset
;
1890 restart
->futex
.flags
= flags
;
1892 ret
= -ERESTART_RESTARTBLOCK
;
1896 hrtimer_cancel(&to
->timer
);
1897 destroy_hrtimer_on_stack(&to
->timer
);
1903 static long futex_wait_restart(struct restart_block
*restart
)
1905 u32 __user
*uaddr
= restart
->futex
.uaddr
;
1906 ktime_t t
, *tp
= NULL
;
1908 if (restart
->futex
.flags
& FLAGS_HAS_TIMEOUT
) {
1909 t
.tv64
= restart
->futex
.time
;
1912 restart
->fn
= do_no_restart_syscall
;
1914 return (long)futex_wait(uaddr
, restart
->futex
.flags
,
1915 restart
->futex
.val
, tp
, restart
->futex
.bitset
);
1920 * Userspace tried a 0 -> TID atomic transition of the futex value
1921 * and failed. The kernel side here does the whole locking operation:
1922 * if there are waiters then it will block, it does PI, etc. (Due to
1923 * races the kernel might see a 0 value of the futex too.)
1925 static int futex_lock_pi(u32 __user
*uaddr
, unsigned int flags
, int detect
,
1926 ktime_t
*time
, int trylock
)
1928 struct hrtimer_sleeper timeout
, *to
= NULL
;
1929 struct futex_hash_bucket
*hb
;
1930 struct futex_q q
= futex_q_init
;
1933 if (refill_pi_state_cache())
1938 hrtimer_init_on_stack(&to
->timer
, CLOCK_REALTIME
,
1940 hrtimer_init_sleeper(to
, current
);
1941 hrtimer_set_expires(&to
->timer
, *time
);
1945 ret
= get_futex_key(uaddr
, flags
& FLAGS_SHARED
, &q
.key
);
1946 if (unlikely(ret
!= 0))
1950 hb
= queue_lock(&q
);
1952 ret
= futex_lock_pi_atomic(uaddr
, hb
, &q
.key
, &q
.pi_state
, current
, 0);
1953 if (unlikely(ret
)) {
1956 /* We got the lock. */
1958 goto out_unlock_put_key
;
1963 * Task is exiting and we just wait for the
1966 queue_unlock(&q
, hb
);
1967 put_futex_key(&q
.key
);
1971 goto out_unlock_put_key
;
1976 * Only actually queue now that the atomic ops are done:
1980 WARN_ON(!q
.pi_state
);
1982 * Block on the PI mutex:
1985 ret
= rt_mutex_timed_lock(&q
.pi_state
->pi_mutex
, to
, 1);
1987 ret
= rt_mutex_trylock(&q
.pi_state
->pi_mutex
);
1988 /* Fixup the trylock return value: */
1989 ret
= ret
? 0 : -EWOULDBLOCK
;
1992 spin_lock(q
.lock_ptr
);
1994 * Fixup the pi_state owner and possibly acquire the lock if we
1997 res
= fixup_owner(uaddr
, &q
, !ret
);
1999 * If fixup_owner() returned an error, proprogate that. If it acquired
2000 * the lock, clear our -ETIMEDOUT or -EINTR.
2003 ret
= (res
< 0) ? res
: 0;
2006 * If fixup_owner() faulted and was unable to handle the fault, unlock
2007 * it and return the fault to userspace.
2009 if (ret
&& (rt_mutex_owner(&q
.pi_state
->pi_mutex
) == current
))
2010 rt_mutex_unlock(&q
.pi_state
->pi_mutex
);
2012 /* Unqueue and drop the lock */
2018 queue_unlock(&q
, hb
);
2021 put_futex_key(&q
.key
);
2024 destroy_hrtimer_on_stack(&to
->timer
);
2025 return ret
!= -EINTR
? ret
: -ERESTARTNOINTR
;
2028 queue_unlock(&q
, hb
);
2030 ret
= fault_in_user_writeable(uaddr
);
2034 if (!(flags
& FLAGS_SHARED
))
2037 put_futex_key(&q
.key
);
2042 * Userspace attempted a TID -> 0 atomic transition, and failed.
2043 * This is the in-kernel slowpath: we look up the PI state (if any),
2044 * and do the rt-mutex unlock.
2046 static int futex_unlock_pi(u32 __user
*uaddr
, unsigned int flags
)
2048 struct futex_hash_bucket
*hb
;
2049 struct futex_q
*this, *next
;
2051 struct plist_head
*head
;
2052 union futex_key key
= FUTEX_KEY_INIT
;
2056 if (get_user(uval
, uaddr
))
2059 * We release only a lock we actually own:
2061 if ((uval
& FUTEX_TID_MASK
) != task_pid_vnr(current
))
2064 ret
= get_futex_key(uaddr
, flags
& FLAGS_SHARED
, &key
);
2065 if (unlikely(ret
!= 0))
2068 hb
= hash_futex(&key
);
2069 spin_lock(&hb
->lock
);
2072 * To avoid races, try to do the TID -> 0 atomic transition
2073 * again. If it succeeds then we can return without waking
2076 if (!(uval
& FUTEX_OWNER_DIED
))
2077 uval
= cmpxchg_futex_value_locked(uaddr
, task_pid_vnr(current
), 0);
2080 if (unlikely(uval
== -EFAULT
))
2083 * Rare case: we managed to release the lock atomically,
2084 * no need to wake anyone else up:
2086 if (unlikely(uval
== task_pid_vnr(current
)))
2090 * Ok, other tasks may need to be woken up - check waiters
2091 * and do the wakeup if necessary:
2095 plist_for_each_entry_safe(this, next
, head
, list
) {
2096 if (!match_futex (&this->key
, &key
))
2098 ret
= wake_futex_pi(uaddr
, uval
, this);
2100 * The atomic access to the futex value
2101 * generated a pagefault, so retry the
2102 * user-access and the wakeup:
2109 * No waiters - kernel unlocks the futex:
2111 if (!(uval
& FUTEX_OWNER_DIED
)) {
2112 ret
= unlock_futex_pi(uaddr
, uval
);
2118 spin_unlock(&hb
->lock
);
2119 put_futex_key(&key
);
2125 spin_unlock(&hb
->lock
);
2126 put_futex_key(&key
);
2128 ret
= fault_in_user_writeable(uaddr
);
2136 * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex
2137 * @hb: the hash_bucket futex_q was original enqueued on
2138 * @q: the futex_q woken while waiting to be requeued
2139 * @key2: the futex_key of the requeue target futex
2140 * @timeout: the timeout associated with the wait (NULL if none)
2142 * Detect if the task was woken on the initial futex as opposed to the requeue
2143 * target futex. If so, determine if it was a timeout or a signal that caused
2144 * the wakeup and return the appropriate error code to the caller. Must be
2145 * called with the hb lock held.
2148 * 0 - no early wakeup detected
2149 * <0 - -ETIMEDOUT or -ERESTARTNOINTR
2152 int handle_early_requeue_pi_wakeup(struct futex_hash_bucket
*hb
,
2153 struct futex_q
*q
, union futex_key
*key2
,
2154 struct hrtimer_sleeper
*timeout
)
2159 * With the hb lock held, we avoid races while we process the wakeup.
2160 * We only need to hold hb (and not hb2) to ensure atomicity as the
2161 * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb.
2162 * It can't be requeued from uaddr2 to something else since we don't
2163 * support a PI aware source futex for requeue.
2165 if (!match_futex(&q
->key
, key2
)) {
2166 WARN_ON(q
->lock_ptr
&& (&hb
->lock
!= q
->lock_ptr
));
2168 * We were woken prior to requeue by a timeout or a signal.
2169 * Unqueue the futex_q and determine which it was.
2171 plist_del(&q
->list
, &q
->list
.plist
);
2173 /* Handle spurious wakeups gracefully */
2175 if (timeout
&& !timeout
->task
)
2177 else if (signal_pending(current
))
2178 ret
= -ERESTARTNOINTR
;
2184 * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2
2185 * @uaddr: the futex we initially wait on (non-pi)
2186 * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be
2187 * the same type, no requeueing from private to shared, etc.
2188 * @val: the expected value of uaddr
2189 * @abs_time: absolute timeout
2190 * @bitset: 32 bit wakeup bitset set by userspace, defaults to all
2191 * @clockrt: whether to use CLOCK_REALTIME (1) or CLOCK_MONOTONIC (0)
2192 * @uaddr2: the pi futex we will take prior to returning to user-space
2194 * The caller will wait on uaddr and will be requeued by futex_requeue() to
2195 * uaddr2 which must be PI aware. Normal wakeup will wake on uaddr2 and
2196 * complete the acquisition of the rt_mutex prior to returning to userspace.
2197 * This ensures the rt_mutex maintains an owner when it has waiters; without
2198 * one, the pi logic wouldn't know which task to boost/deboost, if there was a
2201 * We call schedule in futex_wait_queue_me() when we enqueue and return there
2202 * via the following:
2203 * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue()
2204 * 2) wakeup on uaddr2 after a requeue
2208 * If 3, cleanup and return -ERESTARTNOINTR.
2210 * If 2, we may then block on trying to take the rt_mutex and return via:
2211 * 5) successful lock
2214 * 8) other lock acquisition failure
2216 * If 6, return -EWOULDBLOCK (restarting the syscall would do the same).
2218 * If 4 or 7, we cleanup and return with -ETIMEDOUT.
2224 static int futex_wait_requeue_pi(u32 __user
*uaddr
, unsigned int flags
,
2225 u32 val
, ktime_t
*abs_time
, u32 bitset
,
2228 struct hrtimer_sleeper timeout
, *to
= NULL
;
2229 struct rt_mutex_waiter rt_waiter
;
2230 struct rt_mutex
*pi_mutex
= NULL
;
2231 struct futex_hash_bucket
*hb
;
2232 union futex_key key2
= FUTEX_KEY_INIT
;
2233 struct futex_q q
= futex_q_init
;
2241 hrtimer_init_on_stack(&to
->timer
, (flags
& FLAGS_CLOCKRT
) ?
2242 CLOCK_REALTIME
: CLOCK_MONOTONIC
,
2244 hrtimer_init_sleeper(to
, current
);
2245 hrtimer_set_expires_range_ns(&to
->timer
, *abs_time
,
2246 current
->timer_slack_ns
);
2250 * The waiter is allocated on our stack, manipulated by the requeue
2251 * code while we sleep on uaddr.
2253 debug_rt_mutex_init_waiter(&rt_waiter
);
2254 rt_waiter
.task
= NULL
;
2256 ret
= get_futex_key(uaddr2
, flags
& FLAGS_SHARED
, &key2
);
2257 if (unlikely(ret
!= 0))
2261 q
.rt_waiter
= &rt_waiter
;
2262 q
.requeue_pi_key
= &key2
;
2265 * Prepare to wait on uaddr. On success, increments q.key (key1) ref
2268 ret
= futex_wait_setup(uaddr
, val
, flags
, &q
, &hb
);
2272 /* Queue the futex_q, drop the hb lock, wait for wakeup. */
2273 futex_wait_queue_me(hb
, &q
, to
);
2275 spin_lock(&hb
->lock
);
2276 ret
= handle_early_requeue_pi_wakeup(hb
, &q
, &key2
, to
);
2277 spin_unlock(&hb
->lock
);
2282 * In order for us to be here, we know our q.key == key2, and since
2283 * we took the hb->lock above, we also know that futex_requeue() has
2284 * completed and we no longer have to concern ourselves with a wakeup
2285 * race with the atomic proxy lock acquisition by the requeue code. The
2286 * futex_requeue dropped our key1 reference and incremented our key2
2290 /* Check if the requeue code acquired the second futex for us. */
2293 * Got the lock. We might not be the anticipated owner if we
2294 * did a lock-steal - fix up the PI-state in that case.
2296 if (q
.pi_state
&& (q
.pi_state
->owner
!= current
)) {
2297 spin_lock(q
.lock_ptr
);
2298 ret
= fixup_pi_state_owner(uaddr2
, &q
, current
);
2299 spin_unlock(q
.lock_ptr
);
2303 * We have been woken up by futex_unlock_pi(), a timeout, or a
2304 * signal. futex_unlock_pi() will not destroy the lock_ptr nor
2307 WARN_ON(!&q
.pi_state
);
2308 pi_mutex
= &q
.pi_state
->pi_mutex
;
2309 ret
= rt_mutex_finish_proxy_lock(pi_mutex
, to
, &rt_waiter
, 1);
2310 debug_rt_mutex_free_waiter(&rt_waiter
);
2312 spin_lock(q
.lock_ptr
);
2314 * Fixup the pi_state owner and possibly acquire the lock if we
2317 res
= fixup_owner(uaddr2
, &q
, !ret
);
2319 * If fixup_owner() returned an error, proprogate that. If it
2320 * acquired the lock, clear -ETIMEDOUT or -EINTR.
2323 ret
= (res
< 0) ? res
: 0;
2325 /* Unqueue and drop the lock. */
2330 * If fixup_pi_state_owner() faulted and was unable to handle the
2331 * fault, unlock the rt_mutex and return the fault to userspace.
2333 if (ret
== -EFAULT
) {
2334 if (rt_mutex_owner(pi_mutex
) == current
)
2335 rt_mutex_unlock(pi_mutex
);
2336 } else if (ret
== -EINTR
) {
2338 * We've already been requeued, but cannot restart by calling
2339 * futex_lock_pi() directly. We could restart this syscall, but
2340 * it would detect that the user space "val" changed and return
2341 * -EWOULDBLOCK. Save the overhead of the restart and return
2342 * -EWOULDBLOCK directly.
2348 put_futex_key(&q
.key
);
2350 put_futex_key(&key2
);
2354 hrtimer_cancel(&to
->timer
);
2355 destroy_hrtimer_on_stack(&to
->timer
);
2361 * Support for robust futexes: the kernel cleans up held futexes at
2364 * Implementation: user-space maintains a per-thread list of locks it
2365 * is holding. Upon do_exit(), the kernel carefully walks this list,
2366 * and marks all locks that are owned by this thread with the
2367 * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is
2368 * always manipulated with the lock held, so the list is private and
2369 * per-thread. Userspace also maintains a per-thread 'list_op_pending'
2370 * field, to allow the kernel to clean up if the thread dies after
2371 * acquiring the lock, but just before it could have added itself to
2372 * the list. There can only be one such pending lock.
2376 * sys_set_robust_list() - Set the robust-futex list head of a task
2377 * @head: pointer to the list-head
2378 * @len: length of the list-head, as userspace expects
2380 SYSCALL_DEFINE2(set_robust_list
, struct robust_list_head __user
*, head
,
2383 if (!futex_cmpxchg_enabled
)
2386 * The kernel knows only one size for now:
2388 if (unlikely(len
!= sizeof(*head
)))
2391 current
->robust_list
= head
;
2397 * sys_get_robust_list() - Get the robust-futex list head of a task
2398 * @pid: pid of the process [zero for current task]
2399 * @head_ptr: pointer to a list-head pointer, the kernel fills it in
2400 * @len_ptr: pointer to a length field, the kernel fills in the header size
2402 SYSCALL_DEFINE3(get_robust_list
, int, pid
,
2403 struct robust_list_head __user
* __user
*, head_ptr
,
2404 size_t __user
*, len_ptr
)
2406 struct robust_list_head __user
*head
;
2408 const struct cred
*cred
= current_cred(), *pcred
;
2410 if (!futex_cmpxchg_enabled
)
2414 head
= current
->robust_list
;
2416 struct task_struct
*p
;
2420 p
= find_task_by_vpid(pid
);
2424 pcred
= __task_cred(p
);
2425 if (cred
->euid
!= pcred
->euid
&&
2426 cred
->euid
!= pcred
->uid
&&
2427 !capable(CAP_SYS_PTRACE
))
2429 head
= p
->robust_list
;
2433 if (put_user(sizeof(*head
), len_ptr
))
2435 return put_user(head
, head_ptr
);
2444 * Process a futex-list entry, check whether it's owned by the
2445 * dying task, and do notification if so:
2447 int handle_futex_death(u32 __user
*uaddr
, struct task_struct
*curr
, int pi
)
2449 u32 uval
, nval
, mval
;
2452 if (get_user(uval
, uaddr
))
2455 if ((uval
& FUTEX_TID_MASK
) == task_pid_vnr(curr
)) {
2457 * Ok, this dying thread is truly holding a futex
2458 * of interest. Set the OWNER_DIED bit atomically
2459 * via cmpxchg, and if the value had FUTEX_WAITERS
2460 * set, wake up a waiter (if any). (We have to do a
2461 * futex_wake() even if OWNER_DIED is already set -
2462 * to handle the rare but possible case of recursive
2463 * thread-death.) The rest of the cleanup is done in
2466 mval
= (uval
& FUTEX_WAITERS
) | FUTEX_OWNER_DIED
;
2467 nval
= futex_atomic_cmpxchg_inatomic(uaddr
, uval
, mval
);
2469 if (nval
== -EFAULT
)
2476 * Wake robust non-PI futexes here. The wakeup of
2477 * PI futexes happens in exit_pi_state():
2479 if (!pi
&& (uval
& FUTEX_WAITERS
))
2480 futex_wake(uaddr
, 1, 1, FUTEX_BITSET_MATCH_ANY
);
2486 * Fetch a robust-list pointer. Bit 0 signals PI futexes:
2488 static inline int fetch_robust_entry(struct robust_list __user
**entry
,
2489 struct robust_list __user
* __user
*head
,
2492 unsigned long uentry
;
2494 if (get_user(uentry
, (unsigned long __user
*)head
))
2497 *entry
= (void __user
*)(uentry
& ~1UL);
2504 * Walk curr->robust_list (very carefully, it's a userspace list!)
2505 * and mark any locks found there dead, and notify any waiters.
2507 * We silently return on any sign of list-walking problem.
2509 void exit_robust_list(struct task_struct
*curr
)
2511 struct robust_list_head __user
*head
= curr
->robust_list
;
2512 struct robust_list __user
*entry
, *next_entry
, *pending
;
2513 unsigned int limit
= ROBUST_LIST_LIMIT
, pi
, pip
;
2514 unsigned int uninitialized_var(next_pi
);
2515 unsigned long futex_offset
;
2518 if (!futex_cmpxchg_enabled
)
2522 * Fetch the list head (which was registered earlier, via
2523 * sys_set_robust_list()):
2525 if (fetch_robust_entry(&entry
, &head
->list
.next
, &pi
))
2528 * Fetch the relative futex offset:
2530 if (get_user(futex_offset
, &head
->futex_offset
))
2533 * Fetch any possibly pending lock-add first, and handle it
2536 if (fetch_robust_entry(&pending
, &head
->list_op_pending
, &pip
))
2539 next_entry
= NULL
; /* avoid warning with gcc */
2540 while (entry
!= &head
->list
) {
2542 * Fetch the next entry in the list before calling
2543 * handle_futex_death:
2545 rc
= fetch_robust_entry(&next_entry
, &entry
->next
, &next_pi
);
2547 * A pending lock might already be on the list, so
2548 * don't process it twice:
2550 if (entry
!= pending
)
2551 if (handle_futex_death((void __user
*)entry
+ futex_offset
,
2559 * Avoid excessively long or circular lists:
2568 handle_futex_death((void __user
*)pending
+ futex_offset
,
2572 long do_futex(u32 __user
*uaddr
, int op
, u32 val
, ktime_t
*timeout
,
2573 u32 __user
*uaddr2
, u32 val2
, u32 val3
)
2575 int ret
= -ENOSYS
, cmd
= op
& FUTEX_CMD_MASK
;
2576 unsigned int flags
= 0;
2578 if (!(op
& FUTEX_PRIVATE_FLAG
))
2579 flags
|= FLAGS_SHARED
;
2581 if (op
& FUTEX_CLOCK_REALTIME
) {
2582 flags
|= FLAGS_CLOCKRT
;
2583 if (cmd
!= FUTEX_WAIT_BITSET
&& cmd
!= FUTEX_WAIT_REQUEUE_PI
)
2589 val3
= FUTEX_BITSET_MATCH_ANY
;
2590 case FUTEX_WAIT_BITSET
:
2591 ret
= futex_wait(uaddr
, flags
, val
, timeout
, val3
);
2594 val3
= FUTEX_BITSET_MATCH_ANY
;
2595 case FUTEX_WAKE_BITSET
:
2596 ret
= futex_wake(uaddr
, flags
, val
, val3
);
2599 ret
= futex_requeue(uaddr
, flags
, uaddr2
, val
, val2
, NULL
, 0);
2601 case FUTEX_CMP_REQUEUE
:
2602 ret
= futex_requeue(uaddr
, flags
, uaddr2
, val
, val2
, &val3
, 0);
2605 ret
= futex_wake_op(uaddr
, flags
, uaddr2
, val
, val2
, val3
);
2608 if (futex_cmpxchg_enabled
)
2609 ret
= futex_lock_pi(uaddr
, flags
, val
, timeout
, 0);
2611 case FUTEX_UNLOCK_PI
:
2612 if (futex_cmpxchg_enabled
)
2613 ret
= futex_unlock_pi(uaddr
, flags
);
2615 case FUTEX_TRYLOCK_PI
:
2616 if (futex_cmpxchg_enabled
)
2617 ret
= futex_lock_pi(uaddr
, flags
, 0, timeout
, 1);
2619 case FUTEX_WAIT_REQUEUE_PI
:
2620 val3
= FUTEX_BITSET_MATCH_ANY
;
2621 ret
= futex_wait_requeue_pi(uaddr
, flags
, val
, timeout
, val3
,
2624 case FUTEX_CMP_REQUEUE_PI
:
2625 ret
= futex_requeue(uaddr
, flags
, uaddr2
, val
, val2
, &val3
, 1);
2634 SYSCALL_DEFINE6(futex
, u32 __user
*, uaddr
, int, op
, u32
, val
,
2635 struct timespec __user
*, utime
, u32 __user
*, uaddr2
,
2639 ktime_t t
, *tp
= NULL
;
2641 int cmd
= op
& FUTEX_CMD_MASK
;
2643 if (utime
&& (cmd
== FUTEX_WAIT
|| cmd
== FUTEX_LOCK_PI
||
2644 cmd
== FUTEX_WAIT_BITSET
||
2645 cmd
== FUTEX_WAIT_REQUEUE_PI
)) {
2646 if (copy_from_user(&ts
, utime
, sizeof(ts
)) != 0)
2648 if (!timespec_valid(&ts
))
2651 t
= timespec_to_ktime(ts
);
2652 if (cmd
== FUTEX_WAIT
)
2653 t
= ktime_add_safe(ktime_get(), t
);
2657 * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*.
2658 * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP.
2660 if (cmd
== FUTEX_REQUEUE
|| cmd
== FUTEX_CMP_REQUEUE
||
2661 cmd
== FUTEX_CMP_REQUEUE_PI
|| cmd
== FUTEX_WAKE_OP
)
2662 val2
= (u32
) (unsigned long) utime
;
2664 return do_futex(uaddr
, op
, val
, tp
, uaddr2
, val2
, val3
);
2667 static int __init
futex_init(void)
2673 * This will fail and we want it. Some arch implementations do
2674 * runtime detection of the futex_atomic_cmpxchg_inatomic()
2675 * functionality. We want to know that before we call in any
2676 * of the complex code paths. Also we want to prevent
2677 * registration of robust lists in that case. NULL is
2678 * guaranteed to fault and we get -EFAULT on functional
2679 * implementation, the non-functional ones will return
2682 curval
= cmpxchg_futex_value_locked(NULL
, 0, 0);
2683 if (curval
== -EFAULT
)
2684 futex_cmpxchg_enabled
= 1;
2686 for (i
= 0; i
< ARRAY_SIZE(futex_queues
); i
++) {
2687 plist_head_init(&futex_queues
[i
].chain
, &futex_queues
[i
].lock
);
2688 spin_lock_init(&futex_queues
[i
].lock
);
2693 __initcall(futex_init
);