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 int cmpxchg_futex_value_locked(u32
*curval
, u32 __user
*uaddr
,
385 u32 uval
, u32 newval
)
390 ret
= futex_atomic_cmpxchg_inatomic(curval
, uaddr
, uval
, newval
);
396 static int get_futex_value_locked(u32
*dest
, u32 __user
*from
)
401 ret
= __copy_from_user_inatomic(dest
, from
, sizeof(u32
));
404 return ret
? -EFAULT
: 0;
411 static int refill_pi_state_cache(void)
413 struct futex_pi_state
*pi_state
;
415 if (likely(current
->pi_state_cache
))
418 pi_state
= kzalloc(sizeof(*pi_state
), GFP_KERNEL
);
423 INIT_LIST_HEAD(&pi_state
->list
);
424 /* pi_mutex gets initialized later */
425 pi_state
->owner
= NULL
;
426 atomic_set(&pi_state
->refcount
, 1);
427 pi_state
->key
= FUTEX_KEY_INIT
;
429 current
->pi_state_cache
= pi_state
;
434 static struct futex_pi_state
* alloc_pi_state(void)
436 struct futex_pi_state
*pi_state
= current
->pi_state_cache
;
439 current
->pi_state_cache
= NULL
;
444 static void free_pi_state(struct futex_pi_state
*pi_state
)
446 if (!atomic_dec_and_test(&pi_state
->refcount
))
450 * If pi_state->owner is NULL, the owner is most probably dying
451 * and has cleaned up the pi_state already
453 if (pi_state
->owner
) {
454 raw_spin_lock_irq(&pi_state
->owner
->pi_lock
);
455 list_del_init(&pi_state
->list
);
456 raw_spin_unlock_irq(&pi_state
->owner
->pi_lock
);
458 rt_mutex_proxy_unlock(&pi_state
->pi_mutex
, pi_state
->owner
);
461 if (current
->pi_state_cache
)
465 * pi_state->list is already empty.
466 * clear pi_state->owner.
467 * refcount is at 0 - put it back to 1.
469 pi_state
->owner
= NULL
;
470 atomic_set(&pi_state
->refcount
, 1);
471 current
->pi_state_cache
= pi_state
;
476 * Look up the task based on what TID userspace gave us.
479 static struct task_struct
* futex_find_get_task(pid_t pid
)
481 struct task_struct
*p
;
484 p
= find_task_by_vpid(pid
);
494 * This task is holding PI mutexes at exit time => bad.
495 * Kernel cleans up PI-state, but userspace is likely hosed.
496 * (Robust-futex cleanup is separate and might save the day for userspace.)
498 void exit_pi_state_list(struct task_struct
*curr
)
500 struct list_head
*next
, *head
= &curr
->pi_state_list
;
501 struct futex_pi_state
*pi_state
;
502 struct futex_hash_bucket
*hb
;
503 union futex_key key
= FUTEX_KEY_INIT
;
505 if (!futex_cmpxchg_enabled
)
508 * We are a ZOMBIE and nobody can enqueue itself on
509 * pi_state_list anymore, but we have to be careful
510 * versus waiters unqueueing themselves:
512 raw_spin_lock_irq(&curr
->pi_lock
);
513 while (!list_empty(head
)) {
516 pi_state
= list_entry(next
, struct futex_pi_state
, list
);
518 hb
= hash_futex(&key
);
519 raw_spin_unlock_irq(&curr
->pi_lock
);
521 spin_lock(&hb
->lock
);
523 raw_spin_lock_irq(&curr
->pi_lock
);
525 * We dropped the pi-lock, so re-check whether this
526 * task still owns the PI-state:
528 if (head
->next
!= next
) {
529 spin_unlock(&hb
->lock
);
533 WARN_ON(pi_state
->owner
!= curr
);
534 WARN_ON(list_empty(&pi_state
->list
));
535 list_del_init(&pi_state
->list
);
536 pi_state
->owner
= NULL
;
537 raw_spin_unlock_irq(&curr
->pi_lock
);
539 rt_mutex_unlock(&pi_state
->pi_mutex
);
541 spin_unlock(&hb
->lock
);
543 raw_spin_lock_irq(&curr
->pi_lock
);
545 raw_spin_unlock_irq(&curr
->pi_lock
);
549 lookup_pi_state(u32 uval
, struct futex_hash_bucket
*hb
,
550 union futex_key
*key
, struct futex_pi_state
**ps
)
552 struct futex_pi_state
*pi_state
= NULL
;
553 struct futex_q
*this, *next
;
554 struct plist_head
*head
;
555 struct task_struct
*p
;
556 pid_t pid
= uval
& FUTEX_TID_MASK
;
560 plist_for_each_entry_safe(this, next
, head
, list
) {
561 if (match_futex(&this->key
, key
)) {
563 * Another waiter already exists - bump up
564 * the refcount and return its pi_state:
566 pi_state
= this->pi_state
;
568 * Userspace might have messed up non-PI and PI futexes
570 if (unlikely(!pi_state
))
573 WARN_ON(!atomic_read(&pi_state
->refcount
));
576 * When pi_state->owner is NULL then the owner died
577 * and another waiter is on the fly. pi_state->owner
578 * is fixed up by the task which acquires
579 * pi_state->rt_mutex.
581 * We do not check for pid == 0 which can happen when
582 * the owner died and robust_list_exit() cleared the
585 if (pid
&& pi_state
->owner
) {
587 * Bail out if user space manipulated the
590 if (pid
!= task_pid_vnr(pi_state
->owner
))
594 atomic_inc(&pi_state
->refcount
);
602 * We are the first waiter - try to look up the real owner and attach
603 * the new pi_state to it, but bail out when TID = 0
607 p
= futex_find_get_task(pid
);
612 * We need to look at the task state flags to figure out,
613 * whether the task is exiting. To protect against the do_exit
614 * change of the task flags, we do this protected by
617 raw_spin_lock_irq(&p
->pi_lock
);
618 if (unlikely(p
->flags
& PF_EXITING
)) {
620 * The task is on the way out. When PF_EXITPIDONE is
621 * set, we know that the task has finished the
624 int ret
= (p
->flags
& PF_EXITPIDONE
) ? -ESRCH
: -EAGAIN
;
626 raw_spin_unlock_irq(&p
->pi_lock
);
631 pi_state
= alloc_pi_state();
634 * Initialize the pi_mutex in locked state and make 'p'
637 rt_mutex_init_proxy_locked(&pi_state
->pi_mutex
, p
);
639 /* Store the key for possible exit cleanups: */
640 pi_state
->key
= *key
;
642 WARN_ON(!list_empty(&pi_state
->list
));
643 list_add(&pi_state
->list
, &p
->pi_state_list
);
645 raw_spin_unlock_irq(&p
->pi_lock
);
655 * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex
656 * @uaddr: the pi futex user address
657 * @hb: the pi futex hash bucket
658 * @key: the futex key associated with uaddr and hb
659 * @ps: the pi_state pointer where we store the result of the
661 * @task: the task to perform the atomic lock work for. This will
662 * be "current" except in the case of requeue pi.
663 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
667 * 1 - acquired the lock
670 * The hb->lock and futex_key refs shall be held by the caller.
672 static int futex_lock_pi_atomic(u32 __user
*uaddr
, struct futex_hash_bucket
*hb
,
673 union futex_key
*key
,
674 struct futex_pi_state
**ps
,
675 struct task_struct
*task
, int set_waiters
)
677 int lock_taken
, ret
, ownerdied
= 0;
678 u32 uval
, newval
, curval
, vpid
= task_pid_vnr(task
);
681 ret
= lock_taken
= 0;
684 * To avoid races, we attempt to take the lock here again
685 * (by doing a 0 -> TID atomic cmpxchg), while holding all
686 * the locks. It will most likely not succeed.
690 newval
|= FUTEX_WAITERS
;
692 if (unlikely(cmpxchg_futex_value_locked(&curval
, uaddr
, 0, newval
)))
698 if ((unlikely((curval
& FUTEX_TID_MASK
) == vpid
)))
702 * Surprise - we got the lock. Just return to userspace:
704 if (unlikely(!curval
))
710 * Set the FUTEX_WAITERS flag, so the owner will know it has someone
711 * to wake at the next unlock.
713 newval
= curval
| FUTEX_WAITERS
;
716 * There are two cases, where a futex might have no owner (the
717 * owner TID is 0): OWNER_DIED. We take over the futex in this
718 * case. We also do an unconditional take over, when the owner
721 * This is safe as we are protected by the hash bucket lock !
723 if (unlikely(ownerdied
|| !(curval
& FUTEX_TID_MASK
))) {
724 /* Keep the OWNER_DIED bit */
725 newval
= (curval
& ~FUTEX_TID_MASK
) | vpid
;
730 if (unlikely(cmpxchg_futex_value_locked(&curval
, uaddr
, uval
, newval
)))
732 if (unlikely(curval
!= uval
))
736 * We took the lock due to owner died take over.
738 if (unlikely(lock_taken
))
742 * We dont have the lock. Look up the PI state (or create it if
743 * we are the first waiter):
745 ret
= lookup_pi_state(uval
, hb
, key
, ps
);
751 * No owner found for this futex. Check if the
752 * OWNER_DIED bit is set to figure out whether
753 * this is a robust futex or not.
755 if (get_futex_value_locked(&curval
, uaddr
))
759 * We simply start over in case of a robust
760 * futex. The code above will take the futex
763 if (curval
& FUTEX_OWNER_DIED
) {
776 * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket
777 * @q: The futex_q to unqueue
779 * The q->lock_ptr must not be NULL and must be held by the caller.
781 static void __unqueue_futex(struct futex_q
*q
)
783 struct futex_hash_bucket
*hb
;
785 if (WARN_ON(!q
->lock_ptr
|| !spin_is_locked(q
->lock_ptr
)
786 || plist_node_empty(&q
->list
)))
789 hb
= container_of(q
->lock_ptr
, struct futex_hash_bucket
, lock
);
790 plist_del(&q
->list
, &hb
->chain
);
794 * The hash bucket lock must be held when this is called.
795 * Afterwards, the futex_q must not be accessed.
797 static void wake_futex(struct futex_q
*q
)
799 struct task_struct
*p
= q
->task
;
802 * We set q->lock_ptr = NULL _before_ we wake up the task. If
803 * a non-futex wake up happens on another CPU then the task
804 * might exit and p would dereference a non-existing task
805 * struct. Prevent this by holding a reference on p across the
812 * The waiting task can free the futex_q as soon as
813 * q->lock_ptr = NULL is written, without taking any locks. A
814 * memory barrier is required here to prevent the following
815 * store to lock_ptr from getting ahead of the plist_del.
820 wake_up_state(p
, TASK_NORMAL
);
824 static int wake_futex_pi(u32 __user
*uaddr
, u32 uval
, struct futex_q
*this)
826 struct task_struct
*new_owner
;
827 struct futex_pi_state
*pi_state
= this->pi_state
;
834 * If current does not own the pi_state then the futex is
835 * inconsistent and user space fiddled with the futex value.
837 if (pi_state
->owner
!= current
)
840 raw_spin_lock(&pi_state
->pi_mutex
.wait_lock
);
841 new_owner
= rt_mutex_next_owner(&pi_state
->pi_mutex
);
844 * It is possible that the next waiter (the one that brought
845 * this owner to the kernel) timed out and is no longer
846 * waiting on the lock.
849 new_owner
= this->task
;
852 * We pass it to the next owner. (The WAITERS bit is always
853 * kept enabled while there is PI state around. We must also
854 * preserve the owner died bit.)
856 if (!(uval
& FUTEX_OWNER_DIED
)) {
859 newval
= FUTEX_WAITERS
| task_pid_vnr(new_owner
);
861 if (cmpxchg_futex_value_locked(&curval
, uaddr
, uval
, newval
))
863 else if (curval
!= uval
)
866 raw_spin_unlock(&pi_state
->pi_mutex
.wait_lock
);
871 raw_spin_lock_irq(&pi_state
->owner
->pi_lock
);
872 WARN_ON(list_empty(&pi_state
->list
));
873 list_del_init(&pi_state
->list
);
874 raw_spin_unlock_irq(&pi_state
->owner
->pi_lock
);
876 raw_spin_lock_irq(&new_owner
->pi_lock
);
877 WARN_ON(!list_empty(&pi_state
->list
));
878 list_add(&pi_state
->list
, &new_owner
->pi_state_list
);
879 pi_state
->owner
= new_owner
;
880 raw_spin_unlock_irq(&new_owner
->pi_lock
);
882 raw_spin_unlock(&pi_state
->pi_mutex
.wait_lock
);
883 rt_mutex_unlock(&pi_state
->pi_mutex
);
888 static int unlock_futex_pi(u32 __user
*uaddr
, u32 uval
)
893 * There is no waiter, so we unlock the futex. The owner died
894 * bit has not to be preserved here. We are the owner:
896 if (cmpxchg_futex_value_locked(&oldval
, uaddr
, uval
, 0))
905 * Express the locking dependencies for lockdep:
908 double_lock_hb(struct futex_hash_bucket
*hb1
, struct futex_hash_bucket
*hb2
)
911 spin_lock(&hb1
->lock
);
913 spin_lock_nested(&hb2
->lock
, SINGLE_DEPTH_NESTING
);
914 } else { /* hb1 > hb2 */
915 spin_lock(&hb2
->lock
);
916 spin_lock_nested(&hb1
->lock
, SINGLE_DEPTH_NESTING
);
921 double_unlock_hb(struct futex_hash_bucket
*hb1
, struct futex_hash_bucket
*hb2
)
923 spin_unlock(&hb1
->lock
);
925 spin_unlock(&hb2
->lock
);
929 * Wake up waiters matching bitset queued on this futex (uaddr).
932 futex_wake(u32 __user
*uaddr
, unsigned int flags
, int nr_wake
, u32 bitset
)
934 struct futex_hash_bucket
*hb
;
935 struct futex_q
*this, *next
;
936 struct plist_head
*head
;
937 union futex_key key
= FUTEX_KEY_INIT
;
943 ret
= get_futex_key(uaddr
, flags
& FLAGS_SHARED
, &key
);
944 if (unlikely(ret
!= 0))
947 hb
= hash_futex(&key
);
948 spin_lock(&hb
->lock
);
951 plist_for_each_entry_safe(this, next
, head
, list
) {
952 if (match_futex (&this->key
, &key
)) {
953 if (this->pi_state
|| this->rt_waiter
) {
958 /* Check if one of the bits is set in both bitsets */
959 if (!(this->bitset
& bitset
))
963 if (++ret
>= nr_wake
)
968 spin_unlock(&hb
->lock
);
975 * Wake up all waiters hashed on the physical page that is mapped
976 * to this virtual address:
979 futex_wake_op(u32 __user
*uaddr1
, unsigned int flags
, u32 __user
*uaddr2
,
980 int nr_wake
, int nr_wake2
, int op
)
982 union futex_key key1
= FUTEX_KEY_INIT
, key2
= FUTEX_KEY_INIT
;
983 struct futex_hash_bucket
*hb1
, *hb2
;
984 struct plist_head
*head
;
985 struct futex_q
*this, *next
;
989 ret
= get_futex_key(uaddr1
, flags
& FLAGS_SHARED
, &key1
);
990 if (unlikely(ret
!= 0))
992 ret
= get_futex_key(uaddr2
, flags
& FLAGS_SHARED
, &key2
);
993 if (unlikely(ret
!= 0))
996 hb1
= hash_futex(&key1
);
997 hb2
= hash_futex(&key2
);
1000 double_lock_hb(hb1
, hb2
);
1001 op_ret
= futex_atomic_op_inuser(op
, uaddr2
);
1002 if (unlikely(op_ret
< 0)) {
1004 double_unlock_hb(hb1
, hb2
);
1008 * we don't get EFAULT from MMU faults if we don't have an MMU,
1009 * but we might get them from range checking
1015 if (unlikely(op_ret
!= -EFAULT
)) {
1020 ret
= fault_in_user_writeable(uaddr2
);
1024 if (!(flags
& FLAGS_SHARED
))
1027 put_futex_key(&key2
);
1028 put_futex_key(&key1
);
1034 plist_for_each_entry_safe(this, next
, head
, list
) {
1035 if (match_futex (&this->key
, &key1
)) {
1037 if (++ret
>= nr_wake
)
1046 plist_for_each_entry_safe(this, next
, head
, list
) {
1047 if (match_futex (&this->key
, &key2
)) {
1049 if (++op_ret
>= nr_wake2
)
1056 double_unlock_hb(hb1
, hb2
);
1058 put_futex_key(&key2
);
1060 put_futex_key(&key1
);
1066 * requeue_futex() - Requeue a futex_q from one hb to another
1067 * @q: the futex_q to requeue
1068 * @hb1: the source hash_bucket
1069 * @hb2: the target hash_bucket
1070 * @key2: the new key for the requeued futex_q
1073 void requeue_futex(struct futex_q
*q
, struct futex_hash_bucket
*hb1
,
1074 struct futex_hash_bucket
*hb2
, union futex_key
*key2
)
1078 * If key1 and key2 hash to the same bucket, no need to
1081 if (likely(&hb1
->chain
!= &hb2
->chain
)) {
1082 plist_del(&q
->list
, &hb1
->chain
);
1083 plist_add(&q
->list
, &hb2
->chain
);
1084 q
->lock_ptr
= &hb2
->lock
;
1086 get_futex_key_refs(key2
);
1091 * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue
1093 * @key: the key of the requeue target futex
1094 * @hb: the hash_bucket of the requeue target futex
1096 * During futex_requeue, with requeue_pi=1, it is possible to acquire the
1097 * target futex if it is uncontended or via a lock steal. Set the futex_q key
1098 * to the requeue target futex so the waiter can detect the wakeup on the right
1099 * futex, but remove it from the hb and NULL the rt_waiter so it can detect
1100 * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock
1101 * to protect access to the pi_state to fixup the owner later. Must be called
1102 * with both q->lock_ptr and hb->lock held.
1105 void requeue_pi_wake_futex(struct futex_q
*q
, union futex_key
*key
,
1106 struct futex_hash_bucket
*hb
)
1108 get_futex_key_refs(key
);
1113 WARN_ON(!q
->rt_waiter
);
1114 q
->rt_waiter
= NULL
;
1116 q
->lock_ptr
= &hb
->lock
;
1118 wake_up_state(q
->task
, TASK_NORMAL
);
1122 * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter
1123 * @pifutex: the user address of the to futex
1124 * @hb1: the from futex hash bucket, must be locked by the caller
1125 * @hb2: the to futex hash bucket, must be locked by the caller
1126 * @key1: the from futex key
1127 * @key2: the to futex key
1128 * @ps: address to store the pi_state pointer
1129 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
1131 * Try and get the lock on behalf of the top waiter if we can do it atomically.
1132 * Wake the top waiter if we succeed. If the caller specified set_waiters,
1133 * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit.
1134 * hb1 and hb2 must be held by the caller.
1137 * 0 - failed to acquire the lock atomicly
1138 * 1 - acquired the lock
1141 static int futex_proxy_trylock_atomic(u32 __user
*pifutex
,
1142 struct futex_hash_bucket
*hb1
,
1143 struct futex_hash_bucket
*hb2
,
1144 union futex_key
*key1
, union futex_key
*key2
,
1145 struct futex_pi_state
**ps
, int set_waiters
)
1147 struct futex_q
*top_waiter
= NULL
;
1151 if (get_futex_value_locked(&curval
, pifutex
))
1155 * Find the top_waiter and determine if there are additional waiters.
1156 * If the caller intends to requeue more than 1 waiter to pifutex,
1157 * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now,
1158 * as we have means to handle the possible fault. If not, don't set
1159 * the bit unecessarily as it will force the subsequent unlock to enter
1162 top_waiter
= futex_top_waiter(hb1
, key1
);
1164 /* There are no waiters, nothing for us to do. */
1168 /* Ensure we requeue to the expected futex. */
1169 if (!match_futex(top_waiter
->requeue_pi_key
, key2
))
1173 * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in
1174 * the contended case or if set_waiters is 1. The pi_state is returned
1175 * in ps in contended cases.
1177 ret
= futex_lock_pi_atomic(pifutex
, hb2
, key2
, ps
, top_waiter
->task
,
1180 requeue_pi_wake_futex(top_waiter
, key2
, hb2
);
1186 * futex_requeue() - Requeue waiters from uaddr1 to uaddr2
1187 * @uaddr1: source futex user address
1188 * @flags: futex flags (FLAGS_SHARED, etc.)
1189 * @uaddr2: target futex user address
1190 * @nr_wake: number of waiters to wake (must be 1 for requeue_pi)
1191 * @nr_requeue: number of waiters to requeue (0-INT_MAX)
1192 * @cmpval: @uaddr1 expected value (or %NULL)
1193 * @requeue_pi: if we are attempting to requeue from a non-pi futex to a
1194 * pi futex (pi to pi requeue is not supported)
1196 * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire
1197 * uaddr2 atomically on behalf of the top waiter.
1200 * >=0 - on success, the number of tasks requeued or woken
1203 static int futex_requeue(u32 __user
*uaddr1
, unsigned int flags
,
1204 u32 __user
*uaddr2
, int nr_wake
, int nr_requeue
,
1205 u32
*cmpval
, int requeue_pi
)
1207 union futex_key key1
= FUTEX_KEY_INIT
, key2
= FUTEX_KEY_INIT
;
1208 int drop_count
= 0, task_count
= 0, ret
;
1209 struct futex_pi_state
*pi_state
= NULL
;
1210 struct futex_hash_bucket
*hb1
, *hb2
;
1211 struct plist_head
*head1
;
1212 struct futex_q
*this, *next
;
1217 * requeue_pi requires a pi_state, try to allocate it now
1218 * without any locks in case it fails.
1220 if (refill_pi_state_cache())
1223 * requeue_pi must wake as many tasks as it can, up to nr_wake
1224 * + nr_requeue, since it acquires the rt_mutex prior to
1225 * returning to userspace, so as to not leave the rt_mutex with
1226 * waiters and no owner. However, second and third wake-ups
1227 * cannot be predicted as they involve race conditions with the
1228 * first wake and a fault while looking up the pi_state. Both
1229 * pthread_cond_signal() and pthread_cond_broadcast() should
1237 if (pi_state
!= NULL
) {
1239 * We will have to lookup the pi_state again, so free this one
1240 * to keep the accounting correct.
1242 free_pi_state(pi_state
);
1246 ret
= get_futex_key(uaddr1
, flags
& FLAGS_SHARED
, &key1
);
1247 if (unlikely(ret
!= 0))
1249 ret
= get_futex_key(uaddr2
, flags
& FLAGS_SHARED
, &key2
);
1250 if (unlikely(ret
!= 0))
1253 hb1
= hash_futex(&key1
);
1254 hb2
= hash_futex(&key2
);
1257 double_lock_hb(hb1
, hb2
);
1259 if (likely(cmpval
!= NULL
)) {
1262 ret
= get_futex_value_locked(&curval
, uaddr1
);
1264 if (unlikely(ret
)) {
1265 double_unlock_hb(hb1
, hb2
);
1267 ret
= get_user(curval
, uaddr1
);
1271 if (!(flags
& FLAGS_SHARED
))
1274 put_futex_key(&key2
);
1275 put_futex_key(&key1
);
1278 if (curval
!= *cmpval
) {
1284 if (requeue_pi
&& (task_count
- nr_wake
< nr_requeue
)) {
1286 * Attempt to acquire uaddr2 and wake the top waiter. If we
1287 * intend to requeue waiters, force setting the FUTEX_WAITERS
1288 * bit. We force this here where we are able to easily handle
1289 * faults rather in the requeue loop below.
1291 ret
= futex_proxy_trylock_atomic(uaddr2
, hb1
, hb2
, &key1
,
1292 &key2
, &pi_state
, nr_requeue
);
1295 * At this point the top_waiter has either taken uaddr2 or is
1296 * waiting on it. If the former, then the pi_state will not
1297 * exist yet, look it up one more time to ensure we have a
1304 ret
= get_futex_value_locked(&curval2
, uaddr2
);
1306 ret
= lookup_pi_state(curval2
, hb2
, &key2
,
1314 double_unlock_hb(hb1
, hb2
);
1315 put_futex_key(&key2
);
1316 put_futex_key(&key1
);
1317 ret
= fault_in_user_writeable(uaddr2
);
1322 /* The owner was exiting, try again. */
1323 double_unlock_hb(hb1
, hb2
);
1324 put_futex_key(&key2
);
1325 put_futex_key(&key1
);
1333 head1
= &hb1
->chain
;
1334 plist_for_each_entry_safe(this, next
, head1
, list
) {
1335 if (task_count
- nr_wake
>= nr_requeue
)
1338 if (!match_futex(&this->key
, &key1
))
1342 * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always
1343 * be paired with each other and no other futex ops.
1345 if ((requeue_pi
&& !this->rt_waiter
) ||
1346 (!requeue_pi
&& this->rt_waiter
)) {
1352 * Wake nr_wake waiters. For requeue_pi, if we acquired the
1353 * lock, we already woke the top_waiter. If not, it will be
1354 * woken by futex_unlock_pi().
1356 if (++task_count
<= nr_wake
&& !requeue_pi
) {
1361 /* Ensure we requeue to the expected futex for requeue_pi. */
1362 if (requeue_pi
&& !match_futex(this->requeue_pi_key
, &key2
)) {
1368 * Requeue nr_requeue waiters and possibly one more in the case
1369 * of requeue_pi if we couldn't acquire the lock atomically.
1372 /* Prepare the waiter to take the rt_mutex. */
1373 atomic_inc(&pi_state
->refcount
);
1374 this->pi_state
= pi_state
;
1375 ret
= rt_mutex_start_proxy_lock(&pi_state
->pi_mutex
,
1379 /* We got the lock. */
1380 requeue_pi_wake_futex(this, &key2
, hb2
);
1385 this->pi_state
= NULL
;
1386 free_pi_state(pi_state
);
1390 requeue_futex(this, hb1
, hb2
, &key2
);
1395 double_unlock_hb(hb1
, hb2
);
1398 * drop_futex_key_refs() must be called outside the spinlocks. During
1399 * the requeue we moved futex_q's from the hash bucket at key1 to the
1400 * one at key2 and updated their key pointer. We no longer need to
1401 * hold the references to key1.
1403 while (--drop_count
>= 0)
1404 drop_futex_key_refs(&key1
);
1407 put_futex_key(&key2
);
1409 put_futex_key(&key1
);
1411 if (pi_state
!= NULL
)
1412 free_pi_state(pi_state
);
1413 return ret
? ret
: task_count
;
1416 /* The key must be already stored in q->key. */
1417 static inline struct futex_hash_bucket
*queue_lock(struct futex_q
*q
)
1418 __acquires(&hb
->lock
)
1420 struct futex_hash_bucket
*hb
;
1422 hb
= hash_futex(&q
->key
);
1423 q
->lock_ptr
= &hb
->lock
;
1425 spin_lock(&hb
->lock
);
1430 queue_unlock(struct futex_q
*q
, struct futex_hash_bucket
*hb
)
1431 __releases(&hb
->lock
)
1433 spin_unlock(&hb
->lock
);
1437 * queue_me() - Enqueue the futex_q on the futex_hash_bucket
1438 * @q: The futex_q to enqueue
1439 * @hb: The destination hash bucket
1441 * The hb->lock must be held by the caller, and is released here. A call to
1442 * queue_me() is typically paired with exactly one call to unqueue_me(). The
1443 * exceptions involve the PI related operations, which may use unqueue_me_pi()
1444 * or nothing if the unqueue is done as part of the wake process and the unqueue
1445 * state is implicit in the state of woken task (see futex_wait_requeue_pi() for
1448 static inline void queue_me(struct futex_q
*q
, struct futex_hash_bucket
*hb
)
1449 __releases(&hb
->lock
)
1454 * The priority used to register this element is
1455 * - either the real thread-priority for the real-time threads
1456 * (i.e. threads with a priority lower than MAX_RT_PRIO)
1457 * - or MAX_RT_PRIO for non-RT threads.
1458 * Thus, all RT-threads are woken first in priority order, and
1459 * the others are woken last, in FIFO order.
1461 prio
= min(current
->normal_prio
, MAX_RT_PRIO
);
1463 plist_node_init(&q
->list
, prio
);
1464 plist_add(&q
->list
, &hb
->chain
);
1466 spin_unlock(&hb
->lock
);
1470 * unqueue_me() - Remove the futex_q from its futex_hash_bucket
1471 * @q: The futex_q to unqueue
1473 * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must
1474 * be paired with exactly one earlier call to queue_me().
1477 * 1 - if the futex_q was still queued (and we removed unqueued it)
1478 * 0 - if the futex_q was already removed by the waking thread
1480 static int unqueue_me(struct futex_q
*q
)
1482 spinlock_t
*lock_ptr
;
1485 /* In the common case we don't take the spinlock, which is nice. */
1487 lock_ptr
= q
->lock_ptr
;
1489 if (lock_ptr
!= NULL
) {
1490 spin_lock(lock_ptr
);
1492 * q->lock_ptr can change between reading it and
1493 * spin_lock(), causing us to take the wrong lock. This
1494 * corrects the race condition.
1496 * Reasoning goes like this: if we have the wrong lock,
1497 * q->lock_ptr must have changed (maybe several times)
1498 * between reading it and the spin_lock(). It can
1499 * change again after the spin_lock() but only if it was
1500 * already changed before the spin_lock(). It cannot,
1501 * however, change back to the original value. Therefore
1502 * we can detect whether we acquired the correct lock.
1504 if (unlikely(lock_ptr
!= q
->lock_ptr
)) {
1505 spin_unlock(lock_ptr
);
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
)
1530 BUG_ON(!q
->pi_state
);
1531 free_pi_state(q
->pi_state
);
1534 spin_unlock(q
->lock_ptr
);
1538 * Fixup the pi_state owner with the new owner.
1540 * Must be called with hash bucket lock held and mm->sem held for non
1543 static int fixup_pi_state_owner(u32 __user
*uaddr
, struct futex_q
*q
,
1544 struct task_struct
*newowner
)
1546 u32 newtid
= task_pid_vnr(newowner
) | FUTEX_WAITERS
;
1547 struct futex_pi_state
*pi_state
= q
->pi_state
;
1548 struct task_struct
*oldowner
= pi_state
->owner
;
1549 u32 uval
, curval
, newval
;
1553 if (!pi_state
->owner
)
1554 newtid
|= FUTEX_OWNER_DIED
;
1557 * We are here either because we stole the rtmutex from the
1558 * pending owner or we are the pending owner which failed to
1559 * get the rtmutex. We have to replace the pending owner TID
1560 * in the user space variable. This must be atomic as we have
1561 * to preserve the owner died bit here.
1563 * Note: We write the user space value _before_ changing the pi_state
1564 * because we can fault here. Imagine swapped out pages or a fork
1565 * that marked all the anonymous memory readonly for cow.
1567 * Modifying pi_state _before_ the user space value would
1568 * leave the pi_state in an inconsistent state when we fault
1569 * here, because we need to drop the hash bucket lock to
1570 * handle the fault. This might be observed in the PID check
1571 * in lookup_pi_state.
1574 if (get_futex_value_locked(&uval
, uaddr
))
1578 newval
= (uval
& FUTEX_OWNER_DIED
) | newtid
;
1580 if (cmpxchg_futex_value_locked(&curval
, uaddr
, uval
, newval
))
1588 * We fixed up user space. Now we need to fix the pi_state
1591 if (pi_state
->owner
!= NULL
) {
1592 raw_spin_lock_irq(&pi_state
->owner
->pi_lock
);
1593 WARN_ON(list_empty(&pi_state
->list
));
1594 list_del_init(&pi_state
->list
);
1595 raw_spin_unlock_irq(&pi_state
->owner
->pi_lock
);
1598 pi_state
->owner
= newowner
;
1600 raw_spin_lock_irq(&newowner
->pi_lock
);
1601 WARN_ON(!list_empty(&pi_state
->list
));
1602 list_add(&pi_state
->list
, &newowner
->pi_state_list
);
1603 raw_spin_unlock_irq(&newowner
->pi_lock
);
1607 * To handle the page fault we need to drop the hash bucket
1608 * lock here. That gives the other task (either the pending
1609 * owner itself or the task which stole the rtmutex) the
1610 * chance to try the fixup of the pi_state. So once we are
1611 * back from handling the fault we need to check the pi_state
1612 * after reacquiring the hash bucket lock and before trying to
1613 * do another fixup. When the fixup has been done already we
1617 spin_unlock(q
->lock_ptr
);
1619 ret
= fault_in_user_writeable(uaddr
);
1621 spin_lock(q
->lock_ptr
);
1624 * Check if someone else fixed it for us:
1626 if (pi_state
->owner
!= oldowner
)
1635 static long futex_wait_restart(struct restart_block
*restart
);
1638 * fixup_owner() - Post lock pi_state and corner case management
1639 * @uaddr: user address of the futex
1640 * @q: futex_q (contains pi_state and access to the rt_mutex)
1641 * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0)
1643 * After attempting to lock an rt_mutex, this function is called to cleanup
1644 * the pi_state owner as well as handle race conditions that may allow us to
1645 * acquire the lock. Must be called with the hb lock held.
1648 * 1 - success, lock taken
1649 * 0 - success, lock not taken
1650 * <0 - on error (-EFAULT)
1652 static int fixup_owner(u32 __user
*uaddr
, struct futex_q
*q
, int locked
)
1654 struct task_struct
*owner
;
1659 * Got the lock. We might not be the anticipated owner if we
1660 * did a lock-steal - fix up the PI-state in that case:
1662 if (q
->pi_state
->owner
!= current
)
1663 ret
= fixup_pi_state_owner(uaddr
, q
, current
);
1668 * Catch the rare case, where the lock was released when we were on the
1669 * way back before we locked the hash bucket.
1671 if (q
->pi_state
->owner
== current
) {
1673 * Try to get the rt_mutex now. This might fail as some other
1674 * task acquired the rt_mutex after we removed ourself from the
1675 * rt_mutex waiters list.
1677 if (rt_mutex_trylock(&q
->pi_state
->pi_mutex
)) {
1683 * pi_state is incorrect, some other task did a lock steal and
1684 * we returned due to timeout or signal without taking the
1685 * rt_mutex. Too late. We can access the rt_mutex_owner without
1686 * locking, as the other task is now blocked on the hash bucket
1687 * lock. Fix the state up.
1689 owner
= rt_mutex_owner(&q
->pi_state
->pi_mutex
);
1690 ret
= fixup_pi_state_owner(uaddr
, q
, owner
);
1695 * Paranoia check. If we did not take the lock, then we should not be
1696 * the owner, nor the pending owner, of the rt_mutex.
1698 if (rt_mutex_owner(&q
->pi_state
->pi_mutex
) == current
)
1699 printk(KERN_ERR
"fixup_owner: ret = %d pi-mutex: %p "
1700 "pi-state %p\n", ret
,
1701 q
->pi_state
->pi_mutex
.owner
,
1702 q
->pi_state
->owner
);
1705 return ret
? ret
: locked
;
1709 * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal
1710 * @hb: the futex hash bucket, must be locked by the caller
1711 * @q: the futex_q to queue up on
1712 * @timeout: the prepared hrtimer_sleeper, or null for no timeout
1714 static void futex_wait_queue_me(struct futex_hash_bucket
*hb
, struct futex_q
*q
,
1715 struct hrtimer_sleeper
*timeout
)
1718 * The task state is guaranteed to be set before another task can
1719 * wake it. set_current_state() is implemented using set_mb() and
1720 * queue_me() calls spin_unlock() upon completion, both serializing
1721 * access to the hash list and forcing another memory barrier.
1723 set_current_state(TASK_INTERRUPTIBLE
);
1728 hrtimer_start_expires(&timeout
->timer
, HRTIMER_MODE_ABS
);
1729 if (!hrtimer_active(&timeout
->timer
))
1730 timeout
->task
= NULL
;
1734 * If we have been removed from the hash list, then another task
1735 * has tried to wake us, and we can skip the call to schedule().
1737 if (likely(!plist_node_empty(&q
->list
))) {
1739 * If the timer has already expired, current will already be
1740 * flagged for rescheduling. Only call schedule if there
1741 * is no timeout, or if it has yet to expire.
1743 if (!timeout
|| timeout
->task
)
1746 __set_current_state(TASK_RUNNING
);
1750 * futex_wait_setup() - Prepare to wait on a futex
1751 * @uaddr: the futex userspace address
1752 * @val: the expected value
1753 * @flags: futex flags (FLAGS_SHARED, etc.)
1754 * @q: the associated futex_q
1755 * @hb: storage for hash_bucket pointer to be returned to caller
1757 * Setup the futex_q and locate the hash_bucket. Get the futex value and
1758 * compare it with the expected value. Handle atomic faults internally.
1759 * Return with the hb lock held and a q.key reference on success, and unlocked
1760 * with no q.key reference on failure.
1763 * 0 - uaddr contains val and hb has been locked
1764 * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlcoked
1766 static int futex_wait_setup(u32 __user
*uaddr
, u32 val
, unsigned int flags
,
1767 struct futex_q
*q
, struct futex_hash_bucket
**hb
)
1773 * Access the page AFTER the hash-bucket is locked.
1774 * Order is important:
1776 * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val);
1777 * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); }
1779 * The basic logical guarantee of a futex is that it blocks ONLY
1780 * if cond(var) is known to be true at the time of blocking, for
1781 * any cond. If we locked the hash-bucket after testing *uaddr, that
1782 * would open a race condition where we could block indefinitely with
1783 * cond(var) false, which would violate the guarantee.
1785 * On the other hand, we insert q and release the hash-bucket only
1786 * after testing *uaddr. This guarantees that futex_wait() will NOT
1787 * absorb a wakeup if *uaddr does not match the desired values
1788 * while the syscall executes.
1791 ret
= get_futex_key(uaddr
, flags
& FLAGS_SHARED
, &q
->key
);
1792 if (unlikely(ret
!= 0))
1796 *hb
= queue_lock(q
);
1798 ret
= get_futex_value_locked(&uval
, uaddr
);
1801 queue_unlock(q
, *hb
);
1803 ret
= get_user(uval
, uaddr
);
1807 if (!(flags
& FLAGS_SHARED
))
1810 put_futex_key(&q
->key
);
1815 queue_unlock(q
, *hb
);
1821 put_futex_key(&q
->key
);
1825 static int futex_wait(u32 __user
*uaddr
, unsigned int flags
, u32 val
,
1826 ktime_t
*abs_time
, u32 bitset
)
1828 struct hrtimer_sleeper timeout
, *to
= NULL
;
1829 struct restart_block
*restart
;
1830 struct futex_hash_bucket
*hb
;
1831 struct futex_q q
= futex_q_init
;
1841 hrtimer_init_on_stack(&to
->timer
, (flags
& FLAGS_CLOCKRT
) ?
1842 CLOCK_REALTIME
: CLOCK_MONOTONIC
,
1844 hrtimer_init_sleeper(to
, current
);
1845 hrtimer_set_expires_range_ns(&to
->timer
, *abs_time
,
1846 current
->timer_slack_ns
);
1851 * Prepare to wait on uaddr. On success, holds hb lock and increments
1854 ret
= futex_wait_setup(uaddr
, val
, flags
, &q
, &hb
);
1858 /* queue_me and wait for wakeup, timeout, or a signal. */
1859 futex_wait_queue_me(hb
, &q
, to
);
1861 /* If we were woken (and unqueued), we succeeded, whatever. */
1863 /* unqueue_me() drops q.key ref */
1864 if (!unqueue_me(&q
))
1867 if (to
&& !to
->task
)
1871 * We expect signal_pending(current), but we might be the
1872 * victim of a spurious wakeup as well.
1874 if (!signal_pending(current
))
1881 restart
= ¤t_thread_info()->restart_block
;
1882 restart
->fn
= futex_wait_restart
;
1883 restart
->futex
.uaddr
= uaddr
;
1884 restart
->futex
.val
= val
;
1885 restart
->futex
.time
= abs_time
->tv64
;
1886 restart
->futex
.bitset
= bitset
;
1887 restart
->futex
.flags
= flags
;
1889 ret
= -ERESTART_RESTARTBLOCK
;
1893 hrtimer_cancel(&to
->timer
);
1894 destroy_hrtimer_on_stack(&to
->timer
);
1900 static long futex_wait_restart(struct restart_block
*restart
)
1902 u32 __user
*uaddr
= restart
->futex
.uaddr
;
1903 ktime_t t
, *tp
= NULL
;
1905 if (restart
->futex
.flags
& FLAGS_HAS_TIMEOUT
) {
1906 t
.tv64
= restart
->futex
.time
;
1909 restart
->fn
= do_no_restart_syscall
;
1911 return (long)futex_wait(uaddr
, restart
->futex
.flags
,
1912 restart
->futex
.val
, tp
, restart
->futex
.bitset
);
1917 * Userspace tried a 0 -> TID atomic transition of the futex value
1918 * and failed. The kernel side here does the whole locking operation:
1919 * if there are waiters then it will block, it does PI, etc. (Due to
1920 * races the kernel might see a 0 value of the futex too.)
1922 static int futex_lock_pi(u32 __user
*uaddr
, unsigned int flags
, int detect
,
1923 ktime_t
*time
, int trylock
)
1925 struct hrtimer_sleeper timeout
, *to
= NULL
;
1926 struct futex_hash_bucket
*hb
;
1927 struct futex_q q
= futex_q_init
;
1930 if (refill_pi_state_cache())
1935 hrtimer_init_on_stack(&to
->timer
, CLOCK_REALTIME
,
1937 hrtimer_init_sleeper(to
, current
);
1938 hrtimer_set_expires(&to
->timer
, *time
);
1942 ret
= get_futex_key(uaddr
, flags
& FLAGS_SHARED
, &q
.key
);
1943 if (unlikely(ret
!= 0))
1947 hb
= queue_lock(&q
);
1949 ret
= futex_lock_pi_atomic(uaddr
, hb
, &q
.key
, &q
.pi_state
, current
, 0);
1950 if (unlikely(ret
)) {
1953 /* We got the lock. */
1955 goto out_unlock_put_key
;
1960 * Task is exiting and we just wait for the
1963 queue_unlock(&q
, hb
);
1964 put_futex_key(&q
.key
);
1968 goto out_unlock_put_key
;
1973 * Only actually queue now that the atomic ops are done:
1977 WARN_ON(!q
.pi_state
);
1979 * Block on the PI mutex:
1982 ret
= rt_mutex_timed_lock(&q
.pi_state
->pi_mutex
, to
, 1);
1984 ret
= rt_mutex_trylock(&q
.pi_state
->pi_mutex
);
1985 /* Fixup the trylock return value: */
1986 ret
= ret
? 0 : -EWOULDBLOCK
;
1989 spin_lock(q
.lock_ptr
);
1991 * Fixup the pi_state owner and possibly acquire the lock if we
1994 res
= fixup_owner(uaddr
, &q
, !ret
);
1996 * If fixup_owner() returned an error, proprogate that. If it acquired
1997 * the lock, clear our -ETIMEDOUT or -EINTR.
2000 ret
= (res
< 0) ? res
: 0;
2003 * If fixup_owner() faulted and was unable to handle the fault, unlock
2004 * it and return the fault to userspace.
2006 if (ret
&& (rt_mutex_owner(&q
.pi_state
->pi_mutex
) == current
))
2007 rt_mutex_unlock(&q
.pi_state
->pi_mutex
);
2009 /* Unqueue and drop the lock */
2015 queue_unlock(&q
, hb
);
2018 put_futex_key(&q
.key
);
2021 destroy_hrtimer_on_stack(&to
->timer
);
2022 return ret
!= -EINTR
? ret
: -ERESTARTNOINTR
;
2025 queue_unlock(&q
, hb
);
2027 ret
= fault_in_user_writeable(uaddr
);
2031 if (!(flags
& FLAGS_SHARED
))
2034 put_futex_key(&q
.key
);
2039 * Userspace attempted a TID -> 0 atomic transition, and failed.
2040 * This is the in-kernel slowpath: we look up the PI state (if any),
2041 * and do the rt-mutex unlock.
2043 static int futex_unlock_pi(u32 __user
*uaddr
, unsigned int flags
)
2045 struct futex_hash_bucket
*hb
;
2046 struct futex_q
*this, *next
;
2047 struct plist_head
*head
;
2048 union futex_key key
= FUTEX_KEY_INIT
;
2049 u32 uval
, vpid
= task_pid_vnr(current
);
2053 if (get_user(uval
, uaddr
))
2056 * We release only a lock we actually own:
2058 if ((uval
& FUTEX_TID_MASK
) != vpid
)
2061 ret
= get_futex_key(uaddr
, flags
& FLAGS_SHARED
, &key
);
2062 if (unlikely(ret
!= 0))
2065 hb
= hash_futex(&key
);
2066 spin_lock(&hb
->lock
);
2069 * To avoid races, try to do the TID -> 0 atomic transition
2070 * again. If it succeeds then we can return without waking
2073 if (!(uval
& FUTEX_OWNER_DIED
) &&
2074 cmpxchg_futex_value_locked(&uval
, uaddr
, vpid
, 0))
2077 * Rare case: we managed to release the lock atomically,
2078 * no need to wake anyone else up:
2080 if (unlikely(uval
== vpid
))
2084 * Ok, other tasks may need to be woken up - check waiters
2085 * and do the wakeup if necessary:
2089 plist_for_each_entry_safe(this, next
, head
, list
) {
2090 if (!match_futex (&this->key
, &key
))
2092 ret
= wake_futex_pi(uaddr
, uval
, this);
2094 * The atomic access to the futex value
2095 * generated a pagefault, so retry the
2096 * user-access and the wakeup:
2103 * No waiters - kernel unlocks the futex:
2105 if (!(uval
& FUTEX_OWNER_DIED
)) {
2106 ret
= unlock_futex_pi(uaddr
, uval
);
2112 spin_unlock(&hb
->lock
);
2113 put_futex_key(&key
);
2119 spin_unlock(&hb
->lock
);
2120 put_futex_key(&key
);
2122 ret
= fault_in_user_writeable(uaddr
);
2130 * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex
2131 * @hb: the hash_bucket futex_q was original enqueued on
2132 * @q: the futex_q woken while waiting to be requeued
2133 * @key2: the futex_key of the requeue target futex
2134 * @timeout: the timeout associated with the wait (NULL if none)
2136 * Detect if the task was woken on the initial futex as opposed to the requeue
2137 * target futex. If so, determine if it was a timeout or a signal that caused
2138 * the wakeup and return the appropriate error code to the caller. Must be
2139 * called with the hb lock held.
2142 * 0 - no early wakeup detected
2143 * <0 - -ETIMEDOUT or -ERESTARTNOINTR
2146 int handle_early_requeue_pi_wakeup(struct futex_hash_bucket
*hb
,
2147 struct futex_q
*q
, union futex_key
*key2
,
2148 struct hrtimer_sleeper
*timeout
)
2153 * With the hb lock held, we avoid races while we process the wakeup.
2154 * We only need to hold hb (and not hb2) to ensure atomicity as the
2155 * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb.
2156 * It can't be requeued from uaddr2 to something else since we don't
2157 * support a PI aware source futex for requeue.
2159 if (!match_futex(&q
->key
, key2
)) {
2160 WARN_ON(q
->lock_ptr
&& (&hb
->lock
!= q
->lock_ptr
));
2162 * We were woken prior to requeue by a timeout or a signal.
2163 * Unqueue the futex_q and determine which it was.
2165 plist_del(&q
->list
, &hb
->chain
);
2167 /* Handle spurious wakeups gracefully */
2169 if (timeout
&& !timeout
->task
)
2171 else if (signal_pending(current
))
2172 ret
= -ERESTARTNOINTR
;
2178 * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2
2179 * @uaddr: the futex we initially wait on (non-pi)
2180 * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be
2181 * the same type, no requeueing from private to shared, etc.
2182 * @val: the expected value of uaddr
2183 * @abs_time: absolute timeout
2184 * @bitset: 32 bit wakeup bitset set by userspace, defaults to all
2185 * @clockrt: whether to use CLOCK_REALTIME (1) or CLOCK_MONOTONIC (0)
2186 * @uaddr2: the pi futex we will take prior to returning to user-space
2188 * The caller will wait on uaddr and will be requeued by futex_requeue() to
2189 * uaddr2 which must be PI aware. Normal wakeup will wake on uaddr2 and
2190 * complete the acquisition of the rt_mutex prior to returning to userspace.
2191 * This ensures the rt_mutex maintains an owner when it has waiters; without
2192 * one, the pi logic wouldn't know which task to boost/deboost, if there was a
2195 * We call schedule in futex_wait_queue_me() when we enqueue and return there
2196 * via the following:
2197 * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue()
2198 * 2) wakeup on uaddr2 after a requeue
2202 * If 3, cleanup and return -ERESTARTNOINTR.
2204 * If 2, we may then block on trying to take the rt_mutex and return via:
2205 * 5) successful lock
2208 * 8) other lock acquisition failure
2210 * If 6, return -EWOULDBLOCK (restarting the syscall would do the same).
2212 * If 4 or 7, we cleanup and return with -ETIMEDOUT.
2218 static int futex_wait_requeue_pi(u32 __user
*uaddr
, unsigned int flags
,
2219 u32 val
, ktime_t
*abs_time
, u32 bitset
,
2222 struct hrtimer_sleeper timeout
, *to
= NULL
;
2223 struct rt_mutex_waiter rt_waiter
;
2224 struct rt_mutex
*pi_mutex
= NULL
;
2225 struct futex_hash_bucket
*hb
;
2226 union futex_key key2
= FUTEX_KEY_INIT
;
2227 struct futex_q q
= futex_q_init
;
2235 hrtimer_init_on_stack(&to
->timer
, (flags
& FLAGS_CLOCKRT
) ?
2236 CLOCK_REALTIME
: CLOCK_MONOTONIC
,
2238 hrtimer_init_sleeper(to
, current
);
2239 hrtimer_set_expires_range_ns(&to
->timer
, *abs_time
,
2240 current
->timer_slack_ns
);
2244 * The waiter is allocated on our stack, manipulated by the requeue
2245 * code while we sleep on uaddr.
2247 debug_rt_mutex_init_waiter(&rt_waiter
);
2248 rt_waiter
.task
= NULL
;
2250 ret
= get_futex_key(uaddr2
, flags
& FLAGS_SHARED
, &key2
);
2251 if (unlikely(ret
!= 0))
2255 q
.rt_waiter
= &rt_waiter
;
2256 q
.requeue_pi_key
= &key2
;
2259 * Prepare to wait on uaddr. On success, increments q.key (key1) ref
2262 ret
= futex_wait_setup(uaddr
, val
, flags
, &q
, &hb
);
2266 /* Queue the futex_q, drop the hb lock, wait for wakeup. */
2267 futex_wait_queue_me(hb
, &q
, to
);
2269 spin_lock(&hb
->lock
);
2270 ret
= handle_early_requeue_pi_wakeup(hb
, &q
, &key2
, to
);
2271 spin_unlock(&hb
->lock
);
2276 * In order for us to be here, we know our q.key == key2, and since
2277 * we took the hb->lock above, we also know that futex_requeue() has
2278 * completed and we no longer have to concern ourselves with a wakeup
2279 * race with the atomic proxy lock acquisition by the requeue code. The
2280 * futex_requeue dropped our key1 reference and incremented our key2
2284 /* Check if the requeue code acquired the second futex for us. */
2287 * Got the lock. We might not be the anticipated owner if we
2288 * did a lock-steal - fix up the PI-state in that case.
2290 if (q
.pi_state
&& (q
.pi_state
->owner
!= current
)) {
2291 spin_lock(q
.lock_ptr
);
2292 ret
= fixup_pi_state_owner(uaddr2
, &q
, current
);
2293 spin_unlock(q
.lock_ptr
);
2297 * We have been woken up by futex_unlock_pi(), a timeout, or a
2298 * signal. futex_unlock_pi() will not destroy the lock_ptr nor
2301 WARN_ON(!&q
.pi_state
);
2302 pi_mutex
= &q
.pi_state
->pi_mutex
;
2303 ret
= rt_mutex_finish_proxy_lock(pi_mutex
, to
, &rt_waiter
, 1);
2304 debug_rt_mutex_free_waiter(&rt_waiter
);
2306 spin_lock(q
.lock_ptr
);
2308 * Fixup the pi_state owner and possibly acquire the lock if we
2311 res
= fixup_owner(uaddr2
, &q
, !ret
);
2313 * If fixup_owner() returned an error, proprogate that. If it
2314 * acquired the lock, clear -ETIMEDOUT or -EINTR.
2317 ret
= (res
< 0) ? res
: 0;
2319 /* Unqueue and drop the lock. */
2324 * If fixup_pi_state_owner() faulted and was unable to handle the
2325 * fault, unlock the rt_mutex and return the fault to userspace.
2327 if (ret
== -EFAULT
) {
2328 if (rt_mutex_owner(pi_mutex
) == current
)
2329 rt_mutex_unlock(pi_mutex
);
2330 } else if (ret
== -EINTR
) {
2332 * We've already been requeued, but cannot restart by calling
2333 * futex_lock_pi() directly. We could restart this syscall, but
2334 * it would detect that the user space "val" changed and return
2335 * -EWOULDBLOCK. Save the overhead of the restart and return
2336 * -EWOULDBLOCK directly.
2342 put_futex_key(&q
.key
);
2344 put_futex_key(&key2
);
2348 hrtimer_cancel(&to
->timer
);
2349 destroy_hrtimer_on_stack(&to
->timer
);
2355 * Support for robust futexes: the kernel cleans up held futexes at
2358 * Implementation: user-space maintains a per-thread list of locks it
2359 * is holding. Upon do_exit(), the kernel carefully walks this list,
2360 * and marks all locks that are owned by this thread with the
2361 * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is
2362 * always manipulated with the lock held, so the list is private and
2363 * per-thread. Userspace also maintains a per-thread 'list_op_pending'
2364 * field, to allow the kernel to clean up if the thread dies after
2365 * acquiring the lock, but just before it could have added itself to
2366 * the list. There can only be one such pending lock.
2370 * sys_set_robust_list() - Set the robust-futex list head of a task
2371 * @head: pointer to the list-head
2372 * @len: length of the list-head, as userspace expects
2374 SYSCALL_DEFINE2(set_robust_list
, struct robust_list_head __user
*, head
,
2377 if (!futex_cmpxchg_enabled
)
2380 * The kernel knows only one size for now:
2382 if (unlikely(len
!= sizeof(*head
)))
2385 current
->robust_list
= head
;
2391 * sys_get_robust_list() - Get the robust-futex list head of a task
2392 * @pid: pid of the process [zero for current task]
2393 * @head_ptr: pointer to a list-head pointer, the kernel fills it in
2394 * @len_ptr: pointer to a length field, the kernel fills in the header size
2396 SYSCALL_DEFINE3(get_robust_list
, int, pid
,
2397 struct robust_list_head __user
* __user
*, head_ptr
,
2398 size_t __user
*, len_ptr
)
2400 struct robust_list_head __user
*head
;
2402 const struct cred
*cred
= current_cred(), *pcred
;
2404 if (!futex_cmpxchg_enabled
)
2408 head
= current
->robust_list
;
2410 struct task_struct
*p
;
2414 p
= find_task_by_vpid(pid
);
2418 pcred
= __task_cred(p
);
2419 if (cred
->euid
!= pcred
->euid
&&
2420 cred
->euid
!= pcred
->uid
&&
2421 !capable(CAP_SYS_PTRACE
))
2423 head
= p
->robust_list
;
2427 if (put_user(sizeof(*head
), len_ptr
))
2429 return put_user(head
, head_ptr
);
2438 * Process a futex-list entry, check whether it's owned by the
2439 * dying task, and do notification if so:
2441 int handle_futex_death(u32 __user
*uaddr
, struct task_struct
*curr
, int pi
)
2443 u32 uval
, nval
, mval
;
2446 if (get_user(uval
, uaddr
))
2449 if ((uval
& FUTEX_TID_MASK
) == task_pid_vnr(curr
)) {
2451 * Ok, this dying thread is truly holding a futex
2452 * of interest. Set the OWNER_DIED bit atomically
2453 * via cmpxchg, and if the value had FUTEX_WAITERS
2454 * set, wake up a waiter (if any). (We have to do a
2455 * futex_wake() even if OWNER_DIED is already set -
2456 * to handle the rare but possible case of recursive
2457 * thread-death.) The rest of the cleanup is done in
2460 mval
= (uval
& FUTEX_WAITERS
) | FUTEX_OWNER_DIED
;
2461 if (futex_atomic_cmpxchg_inatomic(&nval
, uaddr
, uval
, mval
))
2468 * Wake robust non-PI futexes here. The wakeup of
2469 * PI futexes happens in exit_pi_state():
2471 if (!pi
&& (uval
& FUTEX_WAITERS
))
2472 futex_wake(uaddr
, 1, 1, FUTEX_BITSET_MATCH_ANY
);
2478 * Fetch a robust-list pointer. Bit 0 signals PI futexes:
2480 static inline int fetch_robust_entry(struct robust_list __user
**entry
,
2481 struct robust_list __user
* __user
*head
,
2484 unsigned long uentry
;
2486 if (get_user(uentry
, (unsigned long __user
*)head
))
2489 *entry
= (void __user
*)(uentry
& ~1UL);
2496 * Walk curr->robust_list (very carefully, it's a userspace list!)
2497 * and mark any locks found there dead, and notify any waiters.
2499 * We silently return on any sign of list-walking problem.
2501 void exit_robust_list(struct task_struct
*curr
)
2503 struct robust_list_head __user
*head
= curr
->robust_list
;
2504 struct robust_list __user
*entry
, *next_entry
, *pending
;
2505 unsigned int limit
= ROBUST_LIST_LIMIT
, pi
, pip
;
2506 unsigned int uninitialized_var(next_pi
);
2507 unsigned long futex_offset
;
2510 if (!futex_cmpxchg_enabled
)
2514 * Fetch the list head (which was registered earlier, via
2515 * sys_set_robust_list()):
2517 if (fetch_robust_entry(&entry
, &head
->list
.next
, &pi
))
2520 * Fetch the relative futex offset:
2522 if (get_user(futex_offset
, &head
->futex_offset
))
2525 * Fetch any possibly pending lock-add first, and handle it
2528 if (fetch_robust_entry(&pending
, &head
->list_op_pending
, &pip
))
2531 next_entry
= NULL
; /* avoid warning with gcc */
2532 while (entry
!= &head
->list
) {
2534 * Fetch the next entry in the list before calling
2535 * handle_futex_death:
2537 rc
= fetch_robust_entry(&next_entry
, &entry
->next
, &next_pi
);
2539 * A pending lock might already be on the list, so
2540 * don't process it twice:
2542 if (entry
!= pending
)
2543 if (handle_futex_death((void __user
*)entry
+ futex_offset
,
2551 * Avoid excessively long or circular lists:
2560 handle_futex_death((void __user
*)pending
+ futex_offset
,
2564 long do_futex(u32 __user
*uaddr
, int op
, u32 val
, ktime_t
*timeout
,
2565 u32 __user
*uaddr2
, u32 val2
, u32 val3
)
2567 int ret
= -ENOSYS
, cmd
= op
& FUTEX_CMD_MASK
;
2568 unsigned int flags
= 0;
2570 if (!(op
& FUTEX_PRIVATE_FLAG
))
2571 flags
|= FLAGS_SHARED
;
2573 if (op
& FUTEX_CLOCK_REALTIME
) {
2574 flags
|= FLAGS_CLOCKRT
;
2575 if (cmd
!= FUTEX_WAIT_BITSET
&& cmd
!= FUTEX_WAIT_REQUEUE_PI
)
2581 val3
= FUTEX_BITSET_MATCH_ANY
;
2582 case FUTEX_WAIT_BITSET
:
2583 ret
= futex_wait(uaddr
, flags
, val
, timeout
, val3
);
2586 val3
= FUTEX_BITSET_MATCH_ANY
;
2587 case FUTEX_WAKE_BITSET
:
2588 ret
= futex_wake(uaddr
, flags
, val
, val3
);
2591 ret
= futex_requeue(uaddr
, flags
, uaddr2
, val
, val2
, NULL
, 0);
2593 case FUTEX_CMP_REQUEUE
:
2594 ret
= futex_requeue(uaddr
, flags
, uaddr2
, val
, val2
, &val3
, 0);
2597 ret
= futex_wake_op(uaddr
, flags
, uaddr2
, val
, val2
, val3
);
2600 if (futex_cmpxchg_enabled
)
2601 ret
= futex_lock_pi(uaddr
, flags
, val
, timeout
, 0);
2603 case FUTEX_UNLOCK_PI
:
2604 if (futex_cmpxchg_enabled
)
2605 ret
= futex_unlock_pi(uaddr
, flags
);
2607 case FUTEX_TRYLOCK_PI
:
2608 if (futex_cmpxchg_enabled
)
2609 ret
= futex_lock_pi(uaddr
, flags
, 0, timeout
, 1);
2611 case FUTEX_WAIT_REQUEUE_PI
:
2612 val3
= FUTEX_BITSET_MATCH_ANY
;
2613 ret
= futex_wait_requeue_pi(uaddr
, flags
, val
, timeout
, val3
,
2616 case FUTEX_CMP_REQUEUE_PI
:
2617 ret
= futex_requeue(uaddr
, flags
, uaddr2
, val
, val2
, &val3
, 1);
2626 SYSCALL_DEFINE6(futex
, u32 __user
*, uaddr
, int, op
, u32
, val
,
2627 struct timespec __user
*, utime
, u32 __user
*, uaddr2
,
2631 ktime_t t
, *tp
= NULL
;
2633 int cmd
= op
& FUTEX_CMD_MASK
;
2635 if (utime
&& (cmd
== FUTEX_WAIT
|| cmd
== FUTEX_LOCK_PI
||
2636 cmd
== FUTEX_WAIT_BITSET
||
2637 cmd
== FUTEX_WAIT_REQUEUE_PI
)) {
2638 if (copy_from_user(&ts
, utime
, sizeof(ts
)) != 0)
2640 if (!timespec_valid(&ts
))
2643 t
= timespec_to_ktime(ts
);
2644 if (cmd
== FUTEX_WAIT
)
2645 t
= ktime_add_safe(ktime_get(), t
);
2649 * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*.
2650 * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP.
2652 if (cmd
== FUTEX_REQUEUE
|| cmd
== FUTEX_CMP_REQUEUE
||
2653 cmd
== FUTEX_CMP_REQUEUE_PI
|| cmd
== FUTEX_WAKE_OP
)
2654 val2
= (u32
) (unsigned long) utime
;
2656 return do_futex(uaddr
, op
, val
, tp
, uaddr2
, val2
, val3
);
2659 static int __init
futex_init(void)
2665 * This will fail and we want it. Some arch implementations do
2666 * runtime detection of the futex_atomic_cmpxchg_inatomic()
2667 * functionality. We want to know that before we call in any
2668 * of the complex code paths. Also we want to prevent
2669 * registration of robust lists in that case. NULL is
2670 * guaranteed to fault and we get -EFAULT on functional
2671 * implementation, the non-functional ones will return
2674 if (cmpxchg_futex_value_locked(&curval
, NULL
, 0, 0) == -EFAULT
)
2675 futex_cmpxchg_enabled
= 1;
2677 for (i
= 0; i
< ARRAY_SIZE(futex_queues
); i
++) {
2678 plist_head_init(&futex_queues
[i
].chain
, &futex_queues
[i
].lock
);
2679 spin_lock_init(&futex_queues
[i
].lock
);
2684 __initcall(futex_init
);