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/export.h>
59 #include <linux/magic.h>
60 #include <linux/pid.h>
61 #include <linux/nsproxy.h>
62 #include <linux/ptrace.h>
63 #include <linux/hugetlb.h>
65 #include <asm/futex.h>
67 #include "rtmutex_common.h"
69 int __read_mostly futex_cmpxchg_enabled;
71 #define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8)
74 * Futex flags used to encode options to functions and preserve them across
77 #define FLAGS_SHARED 0x01
78 #define FLAGS_CLOCKRT 0x02
79 #define FLAGS_HAS_TIMEOUT 0x04
82 * Priority Inheritance state:
84 struct futex_pi_state {
86 * list of 'owned' pi_state instances - these have to be
87 * cleaned up in do_exit() if the task exits prematurely:
89 struct list_head list;
94 struct rt_mutex pi_mutex;
96 struct task_struct *owner;
103 * struct futex_q - The hashed futex queue entry, one per waiting task
104 * @list: priority-sorted list of tasks waiting on this futex
105 * @task: the task waiting on the futex
106 * @lock_ptr: the hash bucket lock
107 * @key: the key the futex is hashed on
108 * @pi_state: optional priority inheritance state
109 * @rt_waiter: rt_waiter storage for use with requeue_pi
110 * @requeue_pi_key: the requeue_pi target futex key
111 * @bitset: bitset for the optional bitmasked wakeup
113 * We use this hashed waitqueue, instead of a normal wait_queue_t, so
114 * we can wake only the relevant ones (hashed queues may be shared).
116 * A futex_q has a woken state, just like tasks have TASK_RUNNING.
117 * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0.
118 * The order of wakeup is always to make the first condition true, then
121 * PI futexes are typically woken before they are removed from the hash list via
122 * the rt_mutex code. See unqueue_me_pi().
125 struct plist_node list;
127 struct task_struct *task;
128 spinlock_t *lock_ptr;
130 struct futex_pi_state *pi_state;
131 struct rt_mutex_waiter *rt_waiter;
132 union futex_key *requeue_pi_key;
136 static const struct futex_q futex_q_init = {
137 /* list gets initialized in queue_me()*/
138 .key = FUTEX_KEY_INIT,
139 .bitset = FUTEX_BITSET_MATCH_ANY
143 * Hash buckets are shared by all the futex_keys that hash to the same
144 * location. Each key may have multiple futex_q structures, one for each task
145 * waiting on a futex.
147 struct futex_hash_bucket {
149 struct plist_head chain;
152 static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS];
155 * We hash on the keys returned from get_futex_key (see below).
157 static struct futex_hash_bucket *hash_futex(union futex_key *key)
159 u32 hash = jhash2((u32*)&key->both.word,
160 (sizeof(key->both.word)+sizeof(key->both.ptr))/4,
162 return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)];
166 * Return 1 if two futex_keys are equal, 0 otherwise.
168 static inline int match_futex(union futex_key *key1, union futex_key *key2)
171 && key1->both.word == key2->both.word
172 && key1->both.ptr == key2->both.ptr
173 && key1->both.offset == key2->both.offset);
177 * Take a reference to the resource addressed by a key.
178 * Can be called while holding spinlocks.
181 static void get_futex_key_refs(union futex_key *key)
186 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
188 ihold(key->shared.inode);
190 case FUT_OFF_MMSHARED:
191 atomic_inc(&key->private.mm->mm_count);
197 * Drop a reference to the resource addressed by a key.
198 * The hash bucket spinlock must not be held.
200 static void drop_futex_key_refs(union futex_key *key)
202 if (!key->both.ptr) {
203 /* If we're here then we tried to put a key we failed to get */
208 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
210 iput(key->shared.inode);
212 case FUT_OFF_MMSHARED:
213 mmdrop(key->private.mm);
219 * get_futex_key() - Get parameters which are the keys for a futex
220 * @uaddr: virtual address of the futex
221 * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED
222 * @key: address where result is stored.
223 * @rw: mapping needs to be read/write (values: VERIFY_READ,
226 * Returns a negative error code or 0
227 * The key words are stored in *key on success.
229 * For shared mappings, it's (page->index, vma->vm_file->f_path.dentry->d_inode,
230 * offset_within_page). For private mappings, it's (uaddr, current->mm).
231 * We can usually work out the index without swapping in the page.
233 * lock_page() might sleep, the caller should not hold a spinlock.
236 get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key, int rw)
238 unsigned long address = (unsigned long)uaddr;
239 struct mm_struct *mm = current->mm;
240 struct page *page, *page_head;
244 * The futex address must be "naturally" aligned.
246 key->both.offset = address % PAGE_SIZE;
247 if (unlikely((address % sizeof(u32)) != 0))
249 address -= key->both.offset;
252 * PROCESS_PRIVATE futexes are fast.
253 * As the mm cannot disappear under us and the 'key' only needs
254 * virtual address, we dont even have to find the underlying vma.
255 * Note : We do have to check 'uaddr' is a valid user address,
256 * but access_ok() should be faster than find_vma()
259 if (unlikely(!access_ok(VERIFY_WRITE, uaddr, sizeof(u32))))
261 key->private.mm = mm;
262 key->private.address = address;
263 get_futex_key_refs(key);
268 err = get_user_pages_fast(address, 1, 1, &page);
270 * If write access is not required (eg. FUTEX_WAIT), try
271 * and get read-only access.
273 if (err == -EFAULT && rw == VERIFY_READ) {
274 err = get_user_pages_fast(address, 1, 0, &page);
282 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
284 if (unlikely(PageTail(page))) {
286 /* serialize against __split_huge_page_splitting() */
288 if (likely(__get_user_pages_fast(address, 1, !ro, &page) == 1)) {
289 page_head = compound_head(page);
291 * page_head is valid pointer but we must pin
292 * it before taking the PG_lock and/or
293 * PG_compound_lock. The moment we re-enable
294 * irqs __split_huge_page_splitting() can
295 * return and the head page can be freed from
296 * under us. We can't take the PG_lock and/or
297 * PG_compound_lock on a page that could be
298 * freed from under us.
300 if (page != page_head) {
311 page_head = compound_head(page);
312 if (page != page_head) {
318 lock_page(page_head);
321 * If page_head->mapping is NULL, then it cannot be a PageAnon
322 * page; but it might be the ZERO_PAGE or in the gate area or
323 * in a special mapping (all cases which we are happy to fail);
324 * or it may have been a good file page when get_user_pages_fast
325 * found it, but truncated or holepunched or subjected to
326 * invalidate_complete_page2 before we got the page lock (also
327 * cases which we are happy to fail). And we hold a reference,
328 * so refcount care in invalidate_complete_page's remove_mapping
329 * prevents drop_caches from setting mapping to NULL beneath us.
331 * The case we do have to guard against is when memory pressure made
332 * shmem_writepage move it from filecache to swapcache beneath us:
333 * an unlikely race, but we do need to retry for page_head->mapping.
335 if (!page_head->mapping) {
336 int shmem_swizzled = PageSwapCache(page_head);
337 unlock_page(page_head);
345 * Private mappings are handled in a simple way.
347 * NOTE: When userspace waits on a MAP_SHARED mapping, even if
348 * it's a read-only handle, it's expected that futexes attach to
349 * the object not the particular process.
351 if (PageAnon(page_head)) {
353 * A RO anonymous page will never change and thus doesn't make
354 * sense for futex operations.
361 key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */
362 key->private.mm = mm;
363 key->private.address = address;
365 key->both.offset |= FUT_OFF_INODE; /* inode-based key */
366 key->shared.inode = page_head->mapping->host;
367 key->shared.pgoff = basepage_index(page);
370 get_futex_key_refs(key);
373 unlock_page(page_head);
378 static inline void put_futex_key(union futex_key *key)
380 drop_futex_key_refs(key);
384 * fault_in_user_writeable() - Fault in user address and verify RW access
385 * @uaddr: pointer to faulting user space address
387 * Slow path to fixup the fault we just took in the atomic write
390 * We have no generic implementation of a non-destructive write to the
391 * user address. We know that we faulted in the atomic pagefault
392 * disabled section so we can as well avoid the #PF overhead by
393 * calling get_user_pages() right away.
395 static int fault_in_user_writeable(u32 __user *uaddr)
397 struct mm_struct *mm = current->mm;
400 down_read(&mm->mmap_sem);
401 ret = fixup_user_fault(current, mm, (unsigned long)uaddr,
403 up_read(&mm->mmap_sem);
405 return ret < 0 ? ret : 0;
409 * futex_top_waiter() - Return the highest priority waiter on a futex
410 * @hb: the hash bucket the futex_q's reside in
411 * @key: the futex key (to distinguish it from other futex futex_q's)
413 * Must be called with the hb lock held.
415 static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb,
416 union futex_key *key)
418 struct futex_q *this;
420 plist_for_each_entry(this, &hb->chain, list) {
421 if (match_futex(&this->key, key))
427 static int cmpxchg_futex_value_locked(u32 *curval, u32 __user *uaddr,
428 u32 uval, u32 newval)
433 ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval);
439 static int get_futex_value_locked(u32 *dest, u32 __user *from)
444 ret = __copy_from_user_inatomic(dest, from, sizeof(u32));
447 return ret ? -EFAULT : 0;
454 static int refill_pi_state_cache(void)
456 struct futex_pi_state *pi_state;
458 if (likely(current->pi_state_cache))
461 pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL);
466 INIT_LIST_HEAD(&pi_state->list);
467 /* pi_mutex gets initialized later */
468 pi_state->owner = NULL;
469 atomic_set(&pi_state->refcount, 1);
470 pi_state->key = FUTEX_KEY_INIT;
472 current->pi_state_cache = pi_state;
477 static struct futex_pi_state * alloc_pi_state(void)
479 struct futex_pi_state *pi_state = current->pi_state_cache;
482 current->pi_state_cache = NULL;
487 static void free_pi_state(struct futex_pi_state *pi_state)
489 if (!atomic_dec_and_test(&pi_state->refcount))
493 * If pi_state->owner is NULL, the owner is most probably dying
494 * and has cleaned up the pi_state already
496 if (pi_state->owner) {
497 raw_spin_lock_irq(&pi_state->owner->pi_lock);
498 list_del_init(&pi_state->list);
499 raw_spin_unlock_irq(&pi_state->owner->pi_lock);
501 rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner);
504 if (current->pi_state_cache)
508 * pi_state->list is already empty.
509 * clear pi_state->owner.
510 * refcount is at 0 - put it back to 1.
512 pi_state->owner = NULL;
513 atomic_set(&pi_state->refcount, 1);
514 current->pi_state_cache = pi_state;
519 * Look up the task based on what TID userspace gave us.
522 static struct task_struct * futex_find_get_task(pid_t pid)
524 struct task_struct *p;
527 p = find_task_by_vpid(pid);
537 * This task is holding PI mutexes at exit time => bad.
538 * Kernel cleans up PI-state, but userspace is likely hosed.
539 * (Robust-futex cleanup is separate and might save the day for userspace.)
541 void exit_pi_state_list(struct task_struct *curr)
543 struct list_head *next, *head = &curr->pi_state_list;
544 struct futex_pi_state *pi_state;
545 struct futex_hash_bucket *hb;
546 union futex_key key = FUTEX_KEY_INIT;
548 if (!futex_cmpxchg_enabled)
551 * We are a ZOMBIE and nobody can enqueue itself on
552 * pi_state_list anymore, but we have to be careful
553 * versus waiters unqueueing themselves:
555 raw_spin_lock_irq(&curr->pi_lock);
556 while (!list_empty(head)) {
559 pi_state = list_entry(next, struct futex_pi_state, list);
561 hb = hash_futex(&key);
562 raw_spin_unlock_irq(&curr->pi_lock);
564 spin_lock(&hb->lock);
566 raw_spin_lock_irq(&curr->pi_lock);
568 * We dropped the pi-lock, so re-check whether this
569 * task still owns the PI-state:
571 if (head->next != next) {
572 spin_unlock(&hb->lock);
576 WARN_ON(pi_state->owner != curr);
577 WARN_ON(list_empty(&pi_state->list));
578 list_del_init(&pi_state->list);
579 pi_state->owner = NULL;
580 raw_spin_unlock_irq(&curr->pi_lock);
582 rt_mutex_unlock(&pi_state->pi_mutex);
584 spin_unlock(&hb->lock);
586 raw_spin_lock_irq(&curr->pi_lock);
588 raw_spin_unlock_irq(&curr->pi_lock);
592 lookup_pi_state(u32 uval, struct futex_hash_bucket *hb,
593 union futex_key *key, struct futex_pi_state **ps,
594 struct task_struct *task)
596 struct futex_pi_state *pi_state = NULL;
597 struct futex_q *this, *next;
598 struct plist_head *head;
599 struct task_struct *p;
600 pid_t pid = uval & FUTEX_TID_MASK;
604 plist_for_each_entry_safe(this, next, head, list) {
605 if (match_futex(&this->key, key)) {
607 * Another waiter already exists - bump up
608 * the refcount and return its pi_state:
610 pi_state = this->pi_state;
612 * Userspace might have messed up non-PI and PI futexes
614 if (unlikely(!pi_state))
617 WARN_ON(!atomic_read(&pi_state->refcount));
620 * When pi_state->owner is NULL then the owner died
621 * and another waiter is on the fly. pi_state->owner
622 * is fixed up by the task which acquires
623 * pi_state->rt_mutex.
625 * We do not check for pid == 0 which can happen when
626 * the owner died and robust_list_exit() cleared the
629 if (pid && pi_state->owner) {
631 * Bail out if user space manipulated the
634 if (pid != task_pid_vnr(pi_state->owner))
639 * Protect against a corrupted uval. If uval
640 * is 0x80000000 then pid is 0 and the waiter
641 * bit is set. So the deadlock check in the
642 * calling code has failed and we did not fall
643 * into the check above due to !pid.
645 if (task && pi_state->owner == task)
648 atomic_inc(&pi_state->refcount);
656 * We are the first waiter - try to look up the real owner and attach
657 * the new pi_state to it, but bail out when TID = 0
661 p = futex_find_get_task(pid);
671 * We need to look at the task state flags to figure out,
672 * whether the task is exiting. To protect against the do_exit
673 * change of the task flags, we do this protected by
676 raw_spin_lock_irq(&p->pi_lock);
677 if (unlikely(p->flags & PF_EXITING)) {
679 * The task is on the way out. When PF_EXITPIDONE is
680 * set, we know that the task has finished the
683 int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN;
685 raw_spin_unlock_irq(&p->pi_lock);
690 pi_state = alloc_pi_state();
693 * Initialize the pi_mutex in locked state and make 'p'
696 rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p);
698 /* Store the key for possible exit cleanups: */
699 pi_state->key = *key;
701 WARN_ON(!list_empty(&pi_state->list));
702 list_add(&pi_state->list, &p->pi_state_list);
704 raw_spin_unlock_irq(&p->pi_lock);
714 * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex
715 * @uaddr: the pi futex user address
716 * @hb: the pi futex hash bucket
717 * @key: the futex key associated with uaddr and hb
718 * @ps: the pi_state pointer where we store the result of the
720 * @task: the task to perform the atomic lock work for. This will
721 * be "current" except in the case of requeue pi.
722 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
726 * 1 - acquired the lock
729 * The hb->lock and futex_key refs shall be held by the caller.
731 static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb,
732 union futex_key *key,
733 struct futex_pi_state **ps,
734 struct task_struct *task, int set_waiters)
736 int lock_taken, ret, force_take = 0;
737 u32 uval, newval, curval, vpid = task_pid_vnr(task);
740 ret = lock_taken = 0;
743 * To avoid races, we attempt to take the lock here again
744 * (by doing a 0 -> TID atomic cmpxchg), while holding all
745 * the locks. It will most likely not succeed.
749 newval |= FUTEX_WAITERS;
751 if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, 0, newval)))
757 if ((unlikely((curval & FUTEX_TID_MASK) == vpid)))
761 * Surprise - we got the lock, but we do not trust user space at all.
763 if (unlikely(!curval)) {
765 * We verify whether there is kernel state for this
766 * futex. If not, we can safely assume, that the 0 ->
767 * TID transition is correct. If state exists, we do
768 * not bother to fixup the user space state as it was
771 return futex_top_waiter(hb, key) ? -EINVAL : 1;
777 * Set the FUTEX_WAITERS flag, so the owner will know it has someone
778 * to wake at the next unlock.
780 newval = curval | FUTEX_WAITERS;
783 * Should we force take the futex? See below.
785 if (unlikely(force_take)) {
787 * Keep the OWNER_DIED and the WAITERS bit and set the
790 newval = (curval & ~FUTEX_TID_MASK) | vpid;
795 if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)))
797 if (unlikely(curval != uval))
801 * We took the lock due to forced take over.
803 if (unlikely(lock_taken))
807 * We dont have the lock. Look up the PI state (or create it if
808 * we are the first waiter):
810 ret = lookup_pi_state(uval, hb, key, ps, task);
816 * We failed to find an owner for this
817 * futex. So we have no pi_state to block
818 * on. This can happen in two cases:
821 * 2) A stale FUTEX_WAITERS bit
823 * Re-read the futex value.
825 if (get_futex_value_locked(&curval, uaddr))
829 * If the owner died or we have a stale
830 * WAITERS bit the owner TID in the user space
833 if (!(curval & FUTEX_TID_MASK)) {
846 * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket
847 * @q: The futex_q to unqueue
849 * The q->lock_ptr must not be NULL and must be held by the caller.
851 static void __unqueue_futex(struct futex_q *q)
853 struct futex_hash_bucket *hb;
855 if (WARN_ON_SMP(!q->lock_ptr || !spin_is_locked(q->lock_ptr))
856 || WARN_ON(plist_node_empty(&q->list)))
859 hb = container_of(q->lock_ptr, struct futex_hash_bucket, lock);
860 plist_del(&q->list, &hb->chain);
864 * The hash bucket lock must be held when this is called.
865 * Afterwards, the futex_q must not be accessed.
867 static void wake_futex(struct futex_q *q)
869 struct task_struct *p = q->task;
871 if (WARN(q->pi_state || q->rt_waiter, "refusing to wake PI futex\n"))
875 * We set q->lock_ptr = NULL _before_ we wake up the task. If
876 * a non-futex wake up happens on another CPU then the task
877 * might exit and p would dereference a non-existing task
878 * struct. Prevent this by holding a reference on p across the
885 * The waiting task can free the futex_q as soon as
886 * q->lock_ptr = NULL is written, without taking any locks. A
887 * memory barrier is required here to prevent the following
888 * store to lock_ptr from getting ahead of the plist_del.
893 wake_up_state(p, TASK_NORMAL);
897 static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this)
899 struct task_struct *new_owner;
900 struct futex_pi_state *pi_state = this->pi_state;
901 u32 uninitialized_var(curval), newval;
907 * If current does not own the pi_state then the futex is
908 * inconsistent and user space fiddled with the futex value.
910 if (pi_state->owner != current)
913 raw_spin_lock(&pi_state->pi_mutex.wait_lock);
914 new_owner = rt_mutex_next_owner(&pi_state->pi_mutex);
917 * It is possible that the next waiter (the one that brought
918 * this owner to the kernel) timed out and is no longer
919 * waiting on the lock.
922 new_owner = this->task;
925 * We pass it to the next owner. (The WAITERS bit is always
926 * kept enabled while there is PI state around. We must also
927 * preserve the owner died bit.)
929 if (!(uval & FUTEX_OWNER_DIED)) {
932 newval = FUTEX_WAITERS | task_pid_vnr(new_owner);
934 if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))
936 else if (curval != uval)
939 raw_spin_unlock(&pi_state->pi_mutex.wait_lock);
944 raw_spin_lock_irq(&pi_state->owner->pi_lock);
945 WARN_ON(list_empty(&pi_state->list));
946 list_del_init(&pi_state->list);
947 raw_spin_unlock_irq(&pi_state->owner->pi_lock);
949 raw_spin_lock_irq(&new_owner->pi_lock);
950 WARN_ON(!list_empty(&pi_state->list));
951 list_add(&pi_state->list, &new_owner->pi_state_list);
952 pi_state->owner = new_owner;
953 raw_spin_unlock_irq(&new_owner->pi_lock);
955 raw_spin_unlock(&pi_state->pi_mutex.wait_lock);
956 rt_mutex_unlock(&pi_state->pi_mutex);
961 static int unlock_futex_pi(u32 __user *uaddr, u32 uval)
963 u32 uninitialized_var(oldval);
966 * There is no waiter, so we unlock the futex. The owner died
967 * bit has not to be preserved here. We are the owner:
969 if (cmpxchg_futex_value_locked(&oldval, uaddr, uval, 0))
978 * Express the locking dependencies for lockdep:
981 double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
984 spin_lock(&hb1->lock);
986 spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING);
987 } else { /* hb1 > hb2 */
988 spin_lock(&hb2->lock);
989 spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING);
994 double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
996 spin_unlock(&hb1->lock);
998 spin_unlock(&hb2->lock);
1002 * Wake up waiters matching bitset queued on this futex (uaddr).
1005 futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset)
1007 struct futex_hash_bucket *hb;
1008 struct futex_q *this, *next;
1009 struct plist_head *head;
1010 union futex_key key = FUTEX_KEY_INIT;
1016 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_READ);
1017 if (unlikely(ret != 0))
1020 hb = hash_futex(&key);
1021 spin_lock(&hb->lock);
1024 plist_for_each_entry_safe(this, next, head, list) {
1025 if (match_futex (&this->key, &key)) {
1026 if (this->pi_state || this->rt_waiter) {
1031 /* Check if one of the bits is set in both bitsets */
1032 if (!(this->bitset & bitset))
1036 if (++ret >= nr_wake)
1041 spin_unlock(&hb->lock);
1042 put_futex_key(&key);
1048 * Wake up all waiters hashed on the physical page that is mapped
1049 * to this virtual address:
1052 futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2,
1053 int nr_wake, int nr_wake2, int op)
1055 union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
1056 struct futex_hash_bucket *hb1, *hb2;
1057 struct plist_head *head;
1058 struct futex_q *this, *next;
1062 ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ);
1063 if (unlikely(ret != 0))
1065 ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE);
1066 if (unlikely(ret != 0))
1069 hb1 = hash_futex(&key1);
1070 hb2 = hash_futex(&key2);
1073 double_lock_hb(hb1, hb2);
1074 op_ret = futex_atomic_op_inuser(op, uaddr2);
1075 if (unlikely(op_ret < 0)) {
1077 double_unlock_hb(hb1, hb2);
1081 * we don't get EFAULT from MMU faults if we don't have an MMU,
1082 * but we might get them from range checking
1088 if (unlikely(op_ret != -EFAULT)) {
1093 ret = fault_in_user_writeable(uaddr2);
1097 if (!(flags & FLAGS_SHARED))
1100 put_futex_key(&key2);
1101 put_futex_key(&key1);
1107 plist_for_each_entry_safe(this, next, head, list) {
1108 if (match_futex (&this->key, &key1)) {
1109 if (this->pi_state || this->rt_waiter) {
1114 if (++ret >= nr_wake)
1123 plist_for_each_entry_safe(this, next, head, list) {
1124 if (match_futex (&this->key, &key2)) {
1125 if (this->pi_state || this->rt_waiter) {
1130 if (++op_ret >= nr_wake2)
1138 double_unlock_hb(hb1, hb2);
1140 put_futex_key(&key2);
1142 put_futex_key(&key1);
1148 * requeue_futex() - Requeue a futex_q from one hb to another
1149 * @q: the futex_q to requeue
1150 * @hb1: the source hash_bucket
1151 * @hb2: the target hash_bucket
1152 * @key2: the new key for the requeued futex_q
1155 void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1,
1156 struct futex_hash_bucket *hb2, union futex_key *key2)
1160 * If key1 and key2 hash to the same bucket, no need to
1163 if (likely(&hb1->chain != &hb2->chain)) {
1164 plist_del(&q->list, &hb1->chain);
1165 plist_add(&q->list, &hb2->chain);
1166 q->lock_ptr = &hb2->lock;
1168 get_futex_key_refs(key2);
1173 * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue
1175 * @key: the key of the requeue target futex
1176 * @hb: the hash_bucket of the requeue target futex
1178 * During futex_requeue, with requeue_pi=1, it is possible to acquire the
1179 * target futex if it is uncontended or via a lock steal. Set the futex_q key
1180 * to the requeue target futex so the waiter can detect the wakeup on the right
1181 * futex, but remove it from the hb and NULL the rt_waiter so it can detect
1182 * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock
1183 * to protect access to the pi_state to fixup the owner later. Must be called
1184 * with both q->lock_ptr and hb->lock held.
1187 void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key,
1188 struct futex_hash_bucket *hb)
1190 get_futex_key_refs(key);
1195 WARN_ON(!q->rt_waiter);
1196 q->rt_waiter = NULL;
1198 q->lock_ptr = &hb->lock;
1200 wake_up_state(q->task, TASK_NORMAL);
1204 * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter
1205 * @pifutex: the user address of the to futex
1206 * @hb1: the from futex hash bucket, must be locked by the caller
1207 * @hb2: the to futex hash bucket, must be locked by the caller
1208 * @key1: the from futex key
1209 * @key2: the to futex key
1210 * @ps: address to store the pi_state pointer
1211 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
1213 * Try and get the lock on behalf of the top waiter if we can do it atomically.
1214 * Wake the top waiter if we succeed. If the caller specified set_waiters,
1215 * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit.
1216 * hb1 and hb2 must be held by the caller.
1219 * 0 - failed to acquire the lock atomicly
1220 * >0 - acquired the lock, return value is vpid of the top_waiter
1223 static int futex_proxy_trylock_atomic(u32 __user *pifutex,
1224 struct futex_hash_bucket *hb1,
1225 struct futex_hash_bucket *hb2,
1226 union futex_key *key1, union futex_key *key2,
1227 struct futex_pi_state **ps, int set_waiters)
1229 struct futex_q *top_waiter = NULL;
1233 if (get_futex_value_locked(&curval, pifutex))
1237 * Find the top_waiter and determine if there are additional waiters.
1238 * If the caller intends to requeue more than 1 waiter to pifutex,
1239 * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now,
1240 * as we have means to handle the possible fault. If not, don't set
1241 * the bit unecessarily as it will force the subsequent unlock to enter
1244 top_waiter = futex_top_waiter(hb1, key1);
1246 /* There are no waiters, nothing for us to do. */
1250 /* Ensure we requeue to the expected futex. */
1251 if (!match_futex(top_waiter->requeue_pi_key, key2))
1255 * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in
1256 * the contended case or if set_waiters is 1. The pi_state is returned
1257 * in ps in contended cases.
1259 vpid = task_pid_vnr(top_waiter->task);
1260 ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task,
1263 requeue_pi_wake_futex(top_waiter, key2, hb2);
1270 * futex_requeue() - Requeue waiters from uaddr1 to uaddr2
1271 * @uaddr1: source futex user address
1272 * @flags: futex flags (FLAGS_SHARED, etc.)
1273 * @uaddr2: target futex user address
1274 * @nr_wake: number of waiters to wake (must be 1 for requeue_pi)
1275 * @nr_requeue: number of waiters to requeue (0-INT_MAX)
1276 * @cmpval: @uaddr1 expected value (or %NULL)
1277 * @requeue_pi: if we are attempting to requeue from a non-pi futex to a
1278 * pi futex (pi to pi requeue is not supported)
1280 * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire
1281 * uaddr2 atomically on behalf of the top waiter.
1284 * >=0 - on success, the number of tasks requeued or woken
1287 static int futex_requeue(u32 __user *uaddr1, unsigned int flags,
1288 u32 __user *uaddr2, int nr_wake, int nr_requeue,
1289 u32 *cmpval, int requeue_pi)
1291 union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
1292 int drop_count = 0, task_count = 0, ret;
1293 struct futex_pi_state *pi_state = NULL;
1294 struct futex_hash_bucket *hb1, *hb2;
1295 struct plist_head *head1;
1296 struct futex_q *this, *next;
1300 * Requeue PI only works on two distinct uaddrs. This
1301 * check is only valid for private futexes. See below.
1303 if (uaddr1 == uaddr2)
1307 * requeue_pi requires a pi_state, try to allocate it now
1308 * without any locks in case it fails.
1310 if (refill_pi_state_cache())
1313 * requeue_pi must wake as many tasks as it can, up to nr_wake
1314 * + nr_requeue, since it acquires the rt_mutex prior to
1315 * returning to userspace, so as to not leave the rt_mutex with
1316 * waiters and no owner. However, second and third wake-ups
1317 * cannot be predicted as they involve race conditions with the
1318 * first wake and a fault while looking up the pi_state. Both
1319 * pthread_cond_signal() and pthread_cond_broadcast() should
1327 if (pi_state != NULL) {
1329 * We will have to lookup the pi_state again, so free this one
1330 * to keep the accounting correct.
1332 free_pi_state(pi_state);
1336 ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ);
1337 if (unlikely(ret != 0))
1339 ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2,
1340 requeue_pi ? VERIFY_WRITE : VERIFY_READ);
1341 if (unlikely(ret != 0))
1345 * The check above which compares uaddrs is not sufficient for
1346 * shared futexes. We need to compare the keys:
1348 if (requeue_pi && match_futex(&key1, &key2)) {
1353 hb1 = hash_futex(&key1);
1354 hb2 = hash_futex(&key2);
1357 double_lock_hb(hb1, hb2);
1359 if (likely(cmpval != NULL)) {
1362 ret = get_futex_value_locked(&curval, uaddr1);
1364 if (unlikely(ret)) {
1365 double_unlock_hb(hb1, hb2);
1367 ret = get_user(curval, uaddr1);
1371 if (!(flags & FLAGS_SHARED))
1374 put_futex_key(&key2);
1375 put_futex_key(&key1);
1378 if (curval != *cmpval) {
1384 if (requeue_pi && (task_count - nr_wake < nr_requeue)) {
1386 * Attempt to acquire uaddr2 and wake the top waiter. If we
1387 * intend to requeue waiters, force setting the FUTEX_WAITERS
1388 * bit. We force this here where we are able to easily handle
1389 * faults rather in the requeue loop below.
1391 ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1,
1392 &key2, &pi_state, nr_requeue);
1395 * At this point the top_waiter has either taken uaddr2 or is
1396 * waiting on it. If the former, then the pi_state will not
1397 * exist yet, look it up one more time to ensure we have a
1398 * reference to it. If the lock was taken, ret contains the
1399 * vpid of the top waiter task.
1406 * If we acquired the lock, then the user
1407 * space value of uaddr2 should be vpid. It
1408 * cannot be changed by the top waiter as it
1409 * is blocked on hb2 lock if it tries to do
1410 * so. If something fiddled with it behind our
1411 * back the pi state lookup might unearth
1412 * it. So we rather use the known value than
1413 * rereading and handing potential crap to
1416 ret = lookup_pi_state(ret, hb2, &key2, &pi_state, NULL);
1423 double_unlock_hb(hb1, hb2);
1424 put_futex_key(&key2);
1425 put_futex_key(&key1);
1426 ret = fault_in_user_writeable(uaddr2);
1431 /* The owner was exiting, try again. */
1432 double_unlock_hb(hb1, hb2);
1433 put_futex_key(&key2);
1434 put_futex_key(&key1);
1442 head1 = &hb1->chain;
1443 plist_for_each_entry_safe(this, next, head1, list) {
1444 if (task_count - nr_wake >= nr_requeue)
1447 if (!match_futex(&this->key, &key1))
1451 * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always
1452 * be paired with each other and no other futex ops.
1454 * We should never be requeueing a futex_q with a pi_state,
1455 * which is awaiting a futex_unlock_pi().
1457 if ((requeue_pi && !this->rt_waiter) ||
1458 (!requeue_pi && this->rt_waiter) ||
1465 * Wake nr_wake waiters. For requeue_pi, if we acquired the
1466 * lock, we already woke the top_waiter. If not, it will be
1467 * woken by futex_unlock_pi().
1469 if (++task_count <= nr_wake && !requeue_pi) {
1474 /* Ensure we requeue to the expected futex for requeue_pi. */
1475 if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) {
1481 * Requeue nr_requeue waiters and possibly one more in the case
1482 * of requeue_pi if we couldn't acquire the lock atomically.
1485 /* Prepare the waiter to take the rt_mutex. */
1486 atomic_inc(&pi_state->refcount);
1487 this->pi_state = pi_state;
1488 ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex,
1492 /* We got the lock. */
1493 requeue_pi_wake_futex(this, &key2, hb2);
1498 this->pi_state = NULL;
1499 free_pi_state(pi_state);
1503 requeue_futex(this, hb1, hb2, &key2);
1508 double_unlock_hb(hb1, hb2);
1511 * drop_futex_key_refs() must be called outside the spinlocks. During
1512 * the requeue we moved futex_q's from the hash bucket at key1 to the
1513 * one at key2 and updated their key pointer. We no longer need to
1514 * hold the references to key1.
1516 while (--drop_count >= 0)
1517 drop_futex_key_refs(&key1);
1520 put_futex_key(&key2);
1522 put_futex_key(&key1);
1524 if (pi_state != NULL)
1525 free_pi_state(pi_state);
1526 return ret ? ret : task_count;
1529 /* The key must be already stored in q->key. */
1530 static inline struct futex_hash_bucket *queue_lock(struct futex_q *q)
1531 __acquires(&hb->lock)
1533 struct futex_hash_bucket *hb;
1535 hb = hash_futex(&q->key);
1536 q->lock_ptr = &hb->lock;
1538 spin_lock(&hb->lock);
1543 queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb)
1544 __releases(&hb->lock)
1546 spin_unlock(&hb->lock);
1550 * queue_me() - Enqueue the futex_q on the futex_hash_bucket
1551 * @q: The futex_q to enqueue
1552 * @hb: The destination hash bucket
1554 * The hb->lock must be held by the caller, and is released here. A call to
1555 * queue_me() is typically paired with exactly one call to unqueue_me(). The
1556 * exceptions involve the PI related operations, which may use unqueue_me_pi()
1557 * or nothing if the unqueue is done as part of the wake process and the unqueue
1558 * state is implicit in the state of woken task (see futex_wait_requeue_pi() for
1561 static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb)
1562 __releases(&hb->lock)
1567 * The priority used to register this element is
1568 * - either the real thread-priority for the real-time threads
1569 * (i.e. threads with a priority lower than MAX_RT_PRIO)
1570 * - or MAX_RT_PRIO for non-RT threads.
1571 * Thus, all RT-threads are woken first in priority order, and
1572 * the others are woken last, in FIFO order.
1574 prio = min(current->normal_prio, MAX_RT_PRIO);
1576 plist_node_init(&q->list, prio);
1577 plist_add(&q->list, &hb->chain);
1579 spin_unlock(&hb->lock);
1583 * unqueue_me() - Remove the futex_q from its futex_hash_bucket
1584 * @q: The futex_q to unqueue
1586 * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must
1587 * be paired with exactly one earlier call to queue_me().
1590 * 1 - if the futex_q was still queued (and we removed unqueued it)
1591 * 0 - if the futex_q was already removed by the waking thread
1593 static int unqueue_me(struct futex_q *q)
1595 spinlock_t *lock_ptr;
1598 /* In the common case we don't take the spinlock, which is nice. */
1600 lock_ptr = q->lock_ptr;
1602 if (lock_ptr != NULL) {
1603 spin_lock(lock_ptr);
1605 * q->lock_ptr can change between reading it and
1606 * spin_lock(), causing us to take the wrong lock. This
1607 * corrects the race condition.
1609 * Reasoning goes like this: if we have the wrong lock,
1610 * q->lock_ptr must have changed (maybe several times)
1611 * between reading it and the spin_lock(). It can
1612 * change again after the spin_lock() but only if it was
1613 * already changed before the spin_lock(). It cannot,
1614 * however, change back to the original value. Therefore
1615 * we can detect whether we acquired the correct lock.
1617 if (unlikely(lock_ptr != q->lock_ptr)) {
1618 spin_unlock(lock_ptr);
1623 BUG_ON(q->pi_state);
1625 spin_unlock(lock_ptr);
1629 drop_futex_key_refs(&q->key);
1634 * PI futexes can not be requeued and must remove themself from the
1635 * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry
1638 static void unqueue_me_pi(struct futex_q *q)
1639 __releases(q->lock_ptr)
1643 BUG_ON(!q->pi_state);
1644 free_pi_state(q->pi_state);
1647 spin_unlock(q->lock_ptr);
1651 * Fixup the pi_state owner with the new owner.
1653 * Must be called with hash bucket lock held and mm->sem held for non
1656 static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q,
1657 struct task_struct *newowner)
1659 u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS;
1660 struct futex_pi_state *pi_state = q->pi_state;
1661 struct task_struct *oldowner = pi_state->owner;
1662 u32 uval, uninitialized_var(curval), newval;
1666 if (!pi_state->owner)
1667 newtid |= FUTEX_OWNER_DIED;
1670 * We are here either because we stole the rtmutex from the
1671 * previous highest priority waiter or we are the highest priority
1672 * waiter but failed to get the rtmutex the first time.
1673 * We have to replace the newowner TID in the user space variable.
1674 * This must be atomic as we have to preserve the owner died bit here.
1676 * Note: We write the user space value _before_ changing the pi_state
1677 * because we can fault here. Imagine swapped out pages or a fork
1678 * that marked all the anonymous memory readonly for cow.
1680 * Modifying pi_state _before_ the user space value would
1681 * leave the pi_state in an inconsistent state when we fault
1682 * here, because we need to drop the hash bucket lock to
1683 * handle the fault. This might be observed in the PID check
1684 * in lookup_pi_state.
1687 if (get_futex_value_locked(&uval, uaddr))
1691 newval = (uval & FUTEX_OWNER_DIED) | newtid;
1693 if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))
1701 * We fixed up user space. Now we need to fix the pi_state
1704 if (pi_state->owner != NULL) {
1705 raw_spin_lock_irq(&pi_state->owner->pi_lock);
1706 WARN_ON(list_empty(&pi_state->list));
1707 list_del_init(&pi_state->list);
1708 raw_spin_unlock_irq(&pi_state->owner->pi_lock);
1711 pi_state->owner = newowner;
1713 raw_spin_lock_irq(&newowner->pi_lock);
1714 WARN_ON(!list_empty(&pi_state->list));
1715 list_add(&pi_state->list, &newowner->pi_state_list);
1716 raw_spin_unlock_irq(&newowner->pi_lock);
1720 * To handle the page fault we need to drop the hash bucket
1721 * lock here. That gives the other task (either the highest priority
1722 * waiter itself or the task which stole the rtmutex) the
1723 * chance to try the fixup of the pi_state. So once we are
1724 * back from handling the fault we need to check the pi_state
1725 * after reacquiring the hash bucket lock and before trying to
1726 * do another fixup. When the fixup has been done already we
1730 spin_unlock(q->lock_ptr);
1732 ret = fault_in_user_writeable(uaddr);
1734 spin_lock(q->lock_ptr);
1737 * Check if someone else fixed it for us:
1739 if (pi_state->owner != oldowner)
1748 static long futex_wait_restart(struct restart_block *restart);
1751 * fixup_owner() - Post lock pi_state and corner case management
1752 * @uaddr: user address of the futex
1753 * @q: futex_q (contains pi_state and access to the rt_mutex)
1754 * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0)
1756 * After attempting to lock an rt_mutex, this function is called to cleanup
1757 * the pi_state owner as well as handle race conditions that may allow us to
1758 * acquire the lock. Must be called with the hb lock held.
1761 * 1 - success, lock taken
1762 * 0 - success, lock not taken
1763 * <0 - on error (-EFAULT)
1765 static int fixup_owner(u32 __user *uaddr, struct futex_q *q, int locked)
1767 struct task_struct *owner;
1772 * Got the lock. We might not be the anticipated owner if we
1773 * did a lock-steal - fix up the PI-state in that case:
1775 if (q->pi_state->owner != current)
1776 ret = fixup_pi_state_owner(uaddr, q, current);
1781 * Catch the rare case, where the lock was released when we were on the
1782 * way back before we locked the hash bucket.
1784 if (q->pi_state->owner == current) {
1786 * Try to get the rt_mutex now. This might fail as some other
1787 * task acquired the rt_mutex after we removed ourself from the
1788 * rt_mutex waiters list.
1790 if (rt_mutex_trylock(&q->pi_state->pi_mutex)) {
1796 * pi_state is incorrect, some other task did a lock steal and
1797 * we returned due to timeout or signal without taking the
1798 * rt_mutex. Too late.
1800 raw_spin_lock(&q->pi_state->pi_mutex.wait_lock);
1801 owner = rt_mutex_owner(&q->pi_state->pi_mutex);
1803 owner = rt_mutex_next_owner(&q->pi_state->pi_mutex);
1804 raw_spin_unlock(&q->pi_state->pi_mutex.wait_lock);
1805 ret = fixup_pi_state_owner(uaddr, q, owner);
1810 * Paranoia check. If we did not take the lock, then we should not be
1811 * the owner of the rt_mutex.
1813 if (rt_mutex_owner(&q->pi_state->pi_mutex) == current)
1814 printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p "
1815 "pi-state %p\n", ret,
1816 q->pi_state->pi_mutex.owner,
1817 q->pi_state->owner);
1820 return ret ? ret : locked;
1824 * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal
1825 * @hb: the futex hash bucket, must be locked by the caller
1826 * @q: the futex_q to queue up on
1827 * @timeout: the prepared hrtimer_sleeper, or null for no timeout
1829 static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q,
1830 struct hrtimer_sleeper *timeout)
1833 * The task state is guaranteed to be set before another task can
1834 * wake it. set_current_state() is implemented using set_mb() and
1835 * queue_me() calls spin_unlock() upon completion, both serializing
1836 * access to the hash list and forcing another memory barrier.
1838 set_current_state(TASK_INTERRUPTIBLE);
1843 hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS);
1844 if (!hrtimer_active(&timeout->timer))
1845 timeout->task = NULL;
1849 * If we have been removed from the hash list, then another task
1850 * has tried to wake us, and we can skip the call to schedule().
1852 if (likely(!plist_node_empty(&q->list))) {
1854 * If the timer has already expired, current will already be
1855 * flagged for rescheduling. Only call schedule if there
1856 * is no timeout, or if it has yet to expire.
1858 if (!timeout || timeout->task)
1861 __set_current_state(TASK_RUNNING);
1865 * futex_wait_setup() - Prepare to wait on a futex
1866 * @uaddr: the futex userspace address
1867 * @val: the expected value
1868 * @flags: futex flags (FLAGS_SHARED, etc.)
1869 * @q: the associated futex_q
1870 * @hb: storage for hash_bucket pointer to be returned to caller
1872 * Setup the futex_q and locate the hash_bucket. Get the futex value and
1873 * compare it with the expected value. Handle atomic faults internally.
1874 * Return with the hb lock held and a q.key reference on success, and unlocked
1875 * with no q.key reference on failure.
1878 * 0 - uaddr contains val and hb has been locked
1879 * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlocked
1881 static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags,
1882 struct futex_q *q, struct futex_hash_bucket **hb)
1888 * Access the page AFTER the hash-bucket is locked.
1889 * Order is important:
1891 * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val);
1892 * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); }
1894 * The basic logical guarantee of a futex is that it blocks ONLY
1895 * if cond(var) is known to be true at the time of blocking, for
1896 * any cond. If we locked the hash-bucket after testing *uaddr, that
1897 * would open a race condition where we could block indefinitely with
1898 * cond(var) false, which would violate the guarantee.
1900 * On the other hand, we insert q and release the hash-bucket only
1901 * after testing *uaddr. This guarantees that futex_wait() will NOT
1902 * absorb a wakeup if *uaddr does not match the desired values
1903 * while the syscall executes.
1906 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q->key, VERIFY_READ);
1907 if (unlikely(ret != 0))
1911 *hb = queue_lock(q);
1913 ret = get_futex_value_locked(&uval, uaddr);
1916 queue_unlock(q, *hb);
1918 ret = get_user(uval, uaddr);
1922 if (!(flags & FLAGS_SHARED))
1925 put_futex_key(&q->key);
1930 queue_unlock(q, *hb);
1936 put_futex_key(&q->key);
1940 static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val,
1941 ktime_t *abs_time, u32 bitset)
1943 struct hrtimer_sleeper timeout, *to = NULL;
1944 struct restart_block *restart;
1945 struct futex_hash_bucket *hb;
1946 struct futex_q q = futex_q_init;
1956 hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ?
1957 CLOCK_REALTIME : CLOCK_MONOTONIC,
1959 hrtimer_init_sleeper(to, current);
1960 hrtimer_set_expires_range_ns(&to->timer, *abs_time,
1961 current->timer_slack_ns);
1966 * Prepare to wait on uaddr. On success, holds hb lock and increments
1969 ret = futex_wait_setup(uaddr, val, flags, &q, &hb);
1973 /* queue_me and wait for wakeup, timeout, or a signal. */
1974 futex_wait_queue_me(hb, &q, to);
1976 /* If we were woken (and unqueued), we succeeded, whatever. */
1978 /* unqueue_me() drops q.key ref */
1979 if (!unqueue_me(&q))
1982 if (to && !to->task)
1986 * We expect signal_pending(current), but we might be the
1987 * victim of a spurious wakeup as well.
1989 if (!signal_pending(current))
1996 restart = ¤t_thread_info()->restart_block;
1997 restart->fn = futex_wait_restart;
1998 restart->futex.uaddr = uaddr;
1999 restart->futex.val = val;
2000 restart->futex.time = abs_time->tv64;
2001 restart->futex.bitset = bitset;
2002 restart->futex.flags = flags | FLAGS_HAS_TIMEOUT;
2004 ret = -ERESTART_RESTARTBLOCK;
2008 hrtimer_cancel(&to->timer);
2009 destroy_hrtimer_on_stack(&to->timer);
2015 static long futex_wait_restart(struct restart_block *restart)
2017 u32 __user *uaddr = restart->futex.uaddr;
2018 ktime_t t, *tp = NULL;
2020 if (restart->futex.flags & FLAGS_HAS_TIMEOUT) {
2021 t.tv64 = restart->futex.time;
2024 restart->fn = do_no_restart_syscall;
2026 return (long)futex_wait(uaddr, restart->futex.flags,
2027 restart->futex.val, tp, restart->futex.bitset);
2032 * Userspace tried a 0 -> TID atomic transition of the futex value
2033 * and failed. The kernel side here does the whole locking operation:
2034 * if there are waiters then it will block, it does PI, etc. (Due to
2035 * races the kernel might see a 0 value of the futex too.)
2037 static int futex_lock_pi(u32 __user *uaddr, unsigned int flags, int detect,
2038 ktime_t *time, int trylock)
2040 struct hrtimer_sleeper timeout, *to = NULL;
2041 struct futex_hash_bucket *hb;
2042 struct futex_q q = futex_q_init;
2045 if (refill_pi_state_cache())
2050 hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME,
2052 hrtimer_init_sleeper(to, current);
2053 hrtimer_set_expires(&to->timer, *time);
2057 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q.key, VERIFY_WRITE);
2058 if (unlikely(ret != 0))
2062 hb = queue_lock(&q);
2064 ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, 0);
2065 if (unlikely(ret)) {
2068 /* We got the lock. */
2070 goto out_unlock_put_key;
2075 * Task is exiting and we just wait for the
2078 queue_unlock(&q, hb);
2079 put_futex_key(&q.key);
2083 goto out_unlock_put_key;
2088 * Only actually queue now that the atomic ops are done:
2092 WARN_ON(!q.pi_state);
2094 * Block on the PI mutex:
2097 ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1);
2099 ret = rt_mutex_trylock(&q.pi_state->pi_mutex);
2100 /* Fixup the trylock return value: */
2101 ret = ret ? 0 : -EWOULDBLOCK;
2104 spin_lock(q.lock_ptr);
2106 * Fixup the pi_state owner and possibly acquire the lock if we
2109 res = fixup_owner(uaddr, &q, !ret);
2111 * If fixup_owner() returned an error, proprogate that. If it acquired
2112 * the lock, clear our -ETIMEDOUT or -EINTR.
2115 ret = (res < 0) ? res : 0;
2118 * If fixup_owner() faulted and was unable to handle the fault, unlock
2119 * it and return the fault to userspace.
2121 if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current))
2122 rt_mutex_unlock(&q.pi_state->pi_mutex);
2124 /* Unqueue and drop the lock */
2130 queue_unlock(&q, hb);
2133 put_futex_key(&q.key);
2136 destroy_hrtimer_on_stack(&to->timer);
2137 return ret != -EINTR ? ret : -ERESTARTNOINTR;
2140 queue_unlock(&q, hb);
2142 ret = fault_in_user_writeable(uaddr);
2146 if (!(flags & FLAGS_SHARED))
2149 put_futex_key(&q.key);
2154 * Userspace attempted a TID -> 0 atomic transition, and failed.
2155 * This is the in-kernel slowpath: we look up the PI state (if any),
2156 * and do the rt-mutex unlock.
2158 static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags)
2160 struct futex_hash_bucket *hb;
2161 struct futex_q *this, *next;
2162 struct plist_head *head;
2163 union futex_key key = FUTEX_KEY_INIT;
2164 u32 uval, vpid = task_pid_vnr(current);
2168 if (get_user(uval, uaddr))
2171 * We release only a lock we actually own:
2173 if ((uval & FUTEX_TID_MASK) != vpid)
2176 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_WRITE);
2177 if (unlikely(ret != 0))
2180 hb = hash_futex(&key);
2181 spin_lock(&hb->lock);
2184 * To avoid races, try to do the TID -> 0 atomic transition
2185 * again. If it succeeds then we can return without waking
2188 if (!(uval & FUTEX_OWNER_DIED) &&
2189 cmpxchg_futex_value_locked(&uval, uaddr, vpid, 0))
2192 * Rare case: we managed to release the lock atomically,
2193 * no need to wake anyone else up:
2195 if (unlikely(uval == vpid))
2199 * Ok, other tasks may need to be woken up - check waiters
2200 * and do the wakeup if necessary:
2204 plist_for_each_entry_safe(this, next, head, list) {
2205 if (!match_futex (&this->key, &key))
2207 ret = wake_futex_pi(uaddr, uval, this);
2209 * The atomic access to the futex value
2210 * generated a pagefault, so retry the
2211 * user-access and the wakeup:
2218 * No waiters - kernel unlocks the futex:
2220 if (!(uval & FUTEX_OWNER_DIED)) {
2221 ret = unlock_futex_pi(uaddr, uval);
2227 spin_unlock(&hb->lock);
2228 put_futex_key(&key);
2234 spin_unlock(&hb->lock);
2235 put_futex_key(&key);
2237 ret = fault_in_user_writeable(uaddr);
2245 * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex
2246 * @hb: the hash_bucket futex_q was original enqueued on
2247 * @q: the futex_q woken while waiting to be requeued
2248 * @key2: the futex_key of the requeue target futex
2249 * @timeout: the timeout associated with the wait (NULL if none)
2251 * Detect if the task was woken on the initial futex as opposed to the requeue
2252 * target futex. If so, determine if it was a timeout or a signal that caused
2253 * the wakeup and return the appropriate error code to the caller. Must be
2254 * called with the hb lock held.
2257 * 0 - no early wakeup detected
2258 * <0 - -ETIMEDOUT or -ERESTARTNOINTR
2261 int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb,
2262 struct futex_q *q, union futex_key *key2,
2263 struct hrtimer_sleeper *timeout)
2268 * With the hb lock held, we avoid races while we process the wakeup.
2269 * We only need to hold hb (and not hb2) to ensure atomicity as the
2270 * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb.
2271 * It can't be requeued from uaddr2 to something else since we don't
2272 * support a PI aware source futex for requeue.
2274 if (!match_futex(&q->key, key2)) {
2275 WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr));
2277 * We were woken prior to requeue by a timeout or a signal.
2278 * Unqueue the futex_q and determine which it was.
2280 plist_del(&q->list, &hb->chain);
2282 /* Handle spurious wakeups gracefully */
2284 if (timeout && !timeout->task)
2286 else if (signal_pending(current))
2287 ret = -ERESTARTNOINTR;
2293 * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2
2294 * @uaddr: the futex we initially wait on (non-pi)
2295 * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be
2296 * the same type, no requeueing from private to shared, etc.
2297 * @val: the expected value of uaddr
2298 * @abs_time: absolute timeout
2299 * @bitset: 32 bit wakeup bitset set by userspace, defaults to all
2300 * @clockrt: whether to use CLOCK_REALTIME (1) or CLOCK_MONOTONIC (0)
2301 * @uaddr2: the pi futex we will take prior to returning to user-space
2303 * The caller will wait on uaddr and will be requeued by futex_requeue() to
2304 * uaddr2 which must be PI aware and unique from uaddr. Normal wakeup will wake
2305 * on uaddr2 and complete the acquisition of the rt_mutex prior to returning to
2306 * userspace. This ensures the rt_mutex maintains an owner when it has waiters;
2307 * without one, the pi logic would not know which task to boost/deboost, if
2308 * there was a need to.
2310 * We call schedule in futex_wait_queue_me() when we enqueue and return there
2311 * via the following:
2312 * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue()
2313 * 2) wakeup on uaddr2 after a requeue
2317 * If 3, cleanup and return -ERESTARTNOINTR.
2319 * If 2, we may then block on trying to take the rt_mutex and return via:
2320 * 5) successful lock
2323 * 8) other lock acquisition failure
2325 * If 6, return -EWOULDBLOCK (restarting the syscall would do the same).
2327 * If 4 or 7, we cleanup and return with -ETIMEDOUT.
2333 static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
2334 u32 val, ktime_t *abs_time, u32 bitset,
2337 struct hrtimer_sleeper timeout, *to = NULL;
2338 struct rt_mutex_waiter rt_waiter;
2339 struct rt_mutex *pi_mutex = NULL;
2340 struct futex_hash_bucket *hb;
2341 union futex_key key2 = FUTEX_KEY_INIT;
2342 struct futex_q q = futex_q_init;
2345 if (uaddr == uaddr2)
2353 hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ?
2354 CLOCK_REALTIME : CLOCK_MONOTONIC,
2356 hrtimer_init_sleeper(to, current);
2357 hrtimer_set_expires_range_ns(&to->timer, *abs_time,
2358 current->timer_slack_ns);
2362 * The waiter is allocated on our stack, manipulated by the requeue
2363 * code while we sleep on uaddr.
2365 debug_rt_mutex_init_waiter(&rt_waiter);
2366 rt_waiter.task = NULL;
2368 ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE);
2369 if (unlikely(ret != 0))
2373 q.rt_waiter = &rt_waiter;
2374 q.requeue_pi_key = &key2;
2377 * Prepare to wait on uaddr. On success, increments q.key (key1) ref
2380 ret = futex_wait_setup(uaddr, val, flags, &q, &hb);
2385 * The check above which compares uaddrs is not sufficient for
2386 * shared futexes. We need to compare the keys:
2388 if (match_futex(&q.key, &key2)) {
2393 /* Queue the futex_q, drop the hb lock, wait for wakeup. */
2394 futex_wait_queue_me(hb, &q, to);
2396 spin_lock(&hb->lock);
2397 ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to);
2398 spin_unlock(&hb->lock);
2403 * In order for us to be here, we know our q.key == key2, and since
2404 * we took the hb->lock above, we also know that futex_requeue() has
2405 * completed and we no longer have to concern ourselves with a wakeup
2406 * race with the atomic proxy lock acquisition by the requeue code. The
2407 * futex_requeue dropped our key1 reference and incremented our key2
2411 /* Check if the requeue code acquired the second futex for us. */
2414 * Got the lock. We might not be the anticipated owner if we
2415 * did a lock-steal - fix up the PI-state in that case.
2417 if (q.pi_state && (q.pi_state->owner != current)) {
2418 spin_lock(q.lock_ptr);
2419 ret = fixup_pi_state_owner(uaddr2, &q, current);
2420 spin_unlock(q.lock_ptr);
2424 * We have been woken up by futex_unlock_pi(), a timeout, or a
2425 * signal. futex_unlock_pi() will not destroy the lock_ptr nor
2428 WARN_ON(!q.pi_state);
2429 pi_mutex = &q.pi_state->pi_mutex;
2430 ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter, 1);
2431 debug_rt_mutex_free_waiter(&rt_waiter);
2433 spin_lock(q.lock_ptr);
2435 * Fixup the pi_state owner and possibly acquire the lock if we
2438 res = fixup_owner(uaddr2, &q, !ret);
2440 * If fixup_owner() returned an error, proprogate that. If it
2441 * acquired the lock, clear -ETIMEDOUT or -EINTR.
2444 ret = (res < 0) ? res : 0;
2446 /* Unqueue and drop the lock. */
2451 * If fixup_pi_state_owner() faulted and was unable to handle the
2452 * fault, unlock the rt_mutex and return the fault to userspace.
2454 if (ret == -EFAULT) {
2455 if (pi_mutex && rt_mutex_owner(pi_mutex) == current)
2456 rt_mutex_unlock(pi_mutex);
2457 } else if (ret == -EINTR) {
2459 * We've already been requeued, but cannot restart by calling
2460 * futex_lock_pi() directly. We could restart this syscall, but
2461 * it would detect that the user space "val" changed and return
2462 * -EWOULDBLOCK. Save the overhead of the restart and return
2463 * -EWOULDBLOCK directly.
2469 put_futex_key(&q.key);
2471 put_futex_key(&key2);
2475 hrtimer_cancel(&to->timer);
2476 destroy_hrtimer_on_stack(&to->timer);
2482 * Support for robust futexes: the kernel cleans up held futexes at
2485 * Implementation: user-space maintains a per-thread list of locks it
2486 * is holding. Upon do_exit(), the kernel carefully walks this list,
2487 * and marks all locks that are owned by this thread with the
2488 * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is
2489 * always manipulated with the lock held, so the list is private and
2490 * per-thread. Userspace also maintains a per-thread 'list_op_pending'
2491 * field, to allow the kernel to clean up if the thread dies after
2492 * acquiring the lock, but just before it could have added itself to
2493 * the list. There can only be one such pending lock.
2497 * sys_set_robust_list() - Set the robust-futex list head of a task
2498 * @head: pointer to the list-head
2499 * @len: length of the list-head, as userspace expects
2501 SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head,
2504 if (!futex_cmpxchg_enabled)
2507 * The kernel knows only one size for now:
2509 if (unlikely(len != sizeof(*head)))
2512 current->robust_list = head;
2518 * sys_get_robust_list() - Get the robust-futex list head of a task
2519 * @pid: pid of the process [zero for current task]
2520 * @head_ptr: pointer to a list-head pointer, the kernel fills it in
2521 * @len_ptr: pointer to a length field, the kernel fills in the header size
2523 SYSCALL_DEFINE3(get_robust_list, int, pid,
2524 struct robust_list_head __user * __user *, head_ptr,
2525 size_t __user *, len_ptr)
2527 struct robust_list_head __user *head;
2529 struct task_struct *p;
2531 if (!futex_cmpxchg_enabled)
2540 p = find_task_by_vpid(pid);
2546 if (!ptrace_may_access(p, PTRACE_MODE_READ))
2549 head = p->robust_list;
2552 if (put_user(sizeof(*head), len_ptr))
2554 return put_user(head, head_ptr);
2563 * Process a futex-list entry, check whether it's owned by the
2564 * dying task, and do notification if so:
2566 int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi)
2568 u32 uval, uninitialized_var(nval), mval;
2571 if (get_user(uval, uaddr))
2574 if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) {
2576 * Ok, this dying thread is truly holding a futex
2577 * of interest. Set the OWNER_DIED bit atomically
2578 * via cmpxchg, and if the value had FUTEX_WAITERS
2579 * set, wake up a waiter (if any). (We have to do a
2580 * futex_wake() even if OWNER_DIED is already set -
2581 * to handle the rare but possible case of recursive
2582 * thread-death.) The rest of the cleanup is done in
2585 mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED;
2587 * We are not holding a lock here, but we want to have
2588 * the pagefault_disable/enable() protection because
2589 * we want to handle the fault gracefully. If the
2590 * access fails we try to fault in the futex with R/W
2591 * verification via get_user_pages. get_user() above
2592 * does not guarantee R/W access. If that fails we
2593 * give up and leave the futex locked.
2595 if (cmpxchg_futex_value_locked(&nval, uaddr, uval, mval)) {
2596 if (fault_in_user_writeable(uaddr))
2604 * Wake robust non-PI futexes here. The wakeup of
2605 * PI futexes happens in exit_pi_state():
2607 if (!pi && (uval & FUTEX_WAITERS))
2608 futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY);
2614 * Fetch a robust-list pointer. Bit 0 signals PI futexes:
2616 static inline int fetch_robust_entry(struct robust_list __user **entry,
2617 struct robust_list __user * __user *head,
2620 unsigned long uentry;
2622 if (get_user(uentry, (unsigned long __user *)head))
2625 *entry = (void __user *)(uentry & ~1UL);
2632 * Walk curr->robust_list (very carefully, it's a userspace list!)
2633 * and mark any locks found there dead, and notify any waiters.
2635 * We silently return on any sign of list-walking problem.
2637 void exit_robust_list(struct task_struct *curr)
2639 struct robust_list_head __user *head = curr->robust_list;
2640 struct robust_list __user *entry, *next_entry, *pending;
2641 unsigned int limit = ROBUST_LIST_LIMIT, pi, pip;
2642 unsigned int uninitialized_var(next_pi);
2643 unsigned long futex_offset;
2646 if (!futex_cmpxchg_enabled)
2650 * Fetch the list head (which was registered earlier, via
2651 * sys_set_robust_list()):
2653 if (fetch_robust_entry(&entry, &head->list.next, &pi))
2656 * Fetch the relative futex offset:
2658 if (get_user(futex_offset, &head->futex_offset))
2661 * Fetch any possibly pending lock-add first, and handle it
2664 if (fetch_robust_entry(&pending, &head->list_op_pending, &pip))
2667 next_entry = NULL; /* avoid warning with gcc */
2668 while (entry != &head->list) {
2670 * Fetch the next entry in the list before calling
2671 * handle_futex_death:
2673 rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi);
2675 * A pending lock might already be on the list, so
2676 * don't process it twice:
2678 if (entry != pending)
2679 if (handle_futex_death((void __user *)entry + futex_offset,
2687 * Avoid excessively long or circular lists:
2696 handle_futex_death((void __user *)pending + futex_offset,
2700 long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout,
2701 u32 __user *uaddr2, u32 val2, u32 val3)
2703 int ret = -ENOSYS, cmd = op & FUTEX_CMD_MASK;
2704 unsigned int flags = 0;
2706 if (!(op & FUTEX_PRIVATE_FLAG))
2707 flags |= FLAGS_SHARED;
2709 if (op & FUTEX_CLOCK_REALTIME) {
2710 flags |= FLAGS_CLOCKRT;
2711 if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI)
2717 case FUTEX_UNLOCK_PI:
2718 case FUTEX_TRYLOCK_PI:
2719 case FUTEX_WAIT_REQUEUE_PI:
2720 case FUTEX_CMP_REQUEUE_PI:
2721 if (!futex_cmpxchg_enabled)
2727 val3 = FUTEX_BITSET_MATCH_ANY;
2728 case FUTEX_WAIT_BITSET:
2729 ret = futex_wait(uaddr, flags, val, timeout, val3);
2732 val3 = FUTEX_BITSET_MATCH_ANY;
2733 case FUTEX_WAKE_BITSET:
2734 ret = futex_wake(uaddr, flags, val, val3);
2737 ret = futex_requeue(uaddr, flags, uaddr2, val, val2, NULL, 0);
2739 case FUTEX_CMP_REQUEUE:
2740 ret = futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 0);
2743 ret = futex_wake_op(uaddr, flags, uaddr2, val, val2, val3);
2746 ret = futex_lock_pi(uaddr, flags, val, timeout, 0);
2748 case FUTEX_UNLOCK_PI:
2749 ret = futex_unlock_pi(uaddr, flags);
2751 case FUTEX_TRYLOCK_PI:
2752 ret = futex_lock_pi(uaddr, flags, 0, timeout, 1);
2754 case FUTEX_WAIT_REQUEUE_PI:
2755 val3 = FUTEX_BITSET_MATCH_ANY;
2756 ret = futex_wait_requeue_pi(uaddr, flags, val, timeout, val3,
2759 case FUTEX_CMP_REQUEUE_PI:
2760 ret = futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 1);
2769 SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val,
2770 struct timespec __user *, utime, u32 __user *, uaddr2,
2774 ktime_t t, *tp = NULL;
2776 int cmd = op & FUTEX_CMD_MASK;
2778 if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI ||
2779 cmd == FUTEX_WAIT_BITSET ||
2780 cmd == FUTEX_WAIT_REQUEUE_PI)) {
2781 if (copy_from_user(&ts, utime, sizeof(ts)) != 0)
2783 if (!timespec_valid(&ts))
2786 t = timespec_to_ktime(ts);
2787 if (cmd == FUTEX_WAIT)
2788 t = ktime_add_safe(ktime_get(), t);
2792 * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*.
2793 * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP.
2795 if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE ||
2796 cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP)
2797 val2 = (u32) (unsigned long) utime;
2799 return do_futex(uaddr, op, val, tp, uaddr2, val2, val3);
2802 static int __init futex_init(void)
2808 * This will fail and we want it. Some arch implementations do
2809 * runtime detection of the futex_atomic_cmpxchg_inatomic()
2810 * functionality. We want to know that before we call in any
2811 * of the complex code paths. Also we want to prevent
2812 * registration of robust lists in that case. NULL is
2813 * guaranteed to fault and we get -EFAULT on functional
2814 * implementation, the non-functional ones will return
2817 if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT)
2818 futex_cmpxchg_enabled = 1;
2820 for (i = 0; i < ARRAY_SIZE(futex_queues); i++) {
2821 plist_head_init(&futex_queues[i].chain);
2822 spin_lock_init(&futex_queues[i].lock);
2827 __initcall(futex_init);