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/sched/rt.h>
64 #include <linux/freezer.h>
66 #include <asm/futex.h>
68 #include "rtmutex_common.h"
70 int __read_mostly futex_cmpxchg_enabled;
72 #define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8)
75 * Futex flags used to encode options to functions and preserve them across
78 #define FLAGS_SHARED 0x01
79 #define FLAGS_CLOCKRT 0x02
80 #define FLAGS_HAS_TIMEOUT 0x04
83 * Priority Inheritance state:
85 struct futex_pi_state {
87 * list of 'owned' pi_state instances - these have to be
88 * cleaned up in do_exit() if the task exits prematurely:
90 struct list_head list;
95 struct rt_mutex pi_mutex;
97 struct task_struct *owner;
104 * struct futex_q - The hashed futex queue entry, one per waiting task
105 * @list: priority-sorted list of tasks waiting on this futex
106 * @task: the task waiting on the futex
107 * @lock_ptr: the hash bucket lock
108 * @key: the key the futex is hashed on
109 * @pi_state: optional priority inheritance state
110 * @rt_waiter: rt_waiter storage for use with requeue_pi
111 * @requeue_pi_key: the requeue_pi target futex key
112 * @bitset: bitset for the optional bitmasked wakeup
114 * We use this hashed waitqueue, instead of a normal wait_queue_t, so
115 * we can wake only the relevant ones (hashed queues may be shared).
117 * A futex_q has a woken state, just like tasks have TASK_RUNNING.
118 * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0.
119 * The order of wakeup is always to make the first condition true, then
122 * PI futexes are typically woken before they are removed from the hash list via
123 * the rt_mutex code. See unqueue_me_pi().
126 struct plist_node list;
128 struct task_struct *task;
129 spinlock_t *lock_ptr;
131 struct futex_pi_state *pi_state;
132 struct rt_mutex_waiter *rt_waiter;
133 union futex_key *requeue_pi_key;
137 static const struct futex_q futex_q_init = {
138 /* list gets initialized in queue_me()*/
139 .key = FUTEX_KEY_INIT,
140 .bitset = FUTEX_BITSET_MATCH_ANY
144 * Hash buckets are shared by all the futex_keys that hash to the same
145 * location. Each key may have multiple futex_q structures, one for each task
146 * waiting on a futex.
148 struct futex_hash_bucket {
150 struct plist_head chain;
153 static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS];
156 * We hash on the keys returned from get_futex_key (see below).
158 static struct futex_hash_bucket *hash_futex(union futex_key *key)
160 u32 hash = jhash2((u32*)&key->both.word,
161 (sizeof(key->both.word)+sizeof(key->both.ptr))/4,
163 return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)];
167 * Return 1 if two futex_keys are equal, 0 otherwise.
169 static inline int match_futex(union futex_key *key1, union futex_key *key2)
172 && key1->both.word == key2->both.word
173 && key1->both.ptr == key2->both.ptr
174 && key1->both.offset == key2->both.offset);
178 * Take a reference to the resource addressed by a key.
179 * Can be called while holding spinlocks.
182 static void get_futex_key_refs(union futex_key *key)
187 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
189 ihold(key->shared.inode);
191 case FUT_OFF_MMSHARED:
192 atomic_inc(&key->private.mm->mm_count);
198 * Drop a reference to the resource addressed by a key.
199 * The hash bucket spinlock must not be held.
201 static void drop_futex_key_refs(union futex_key *key)
203 if (!key->both.ptr) {
204 /* If we're here then we tried to put a key we failed to get */
209 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
211 iput(key->shared.inode);
213 case FUT_OFF_MMSHARED:
214 mmdrop(key->private.mm);
220 * get_futex_key() - Get parameters which are the keys for a futex
221 * @uaddr: virtual address of the futex
222 * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED
223 * @key: address where result is stored.
224 * @rw: mapping needs to be read/write (values: VERIFY_READ,
227 * Return: a negative error code or 0
229 * The key words are stored in *key on success.
231 * For shared mappings, it's (page->index, file_inode(vma->vm_file),
232 * offset_within_page). For private mappings, it's (uaddr, current->mm).
233 * We can usually work out the index without swapping in the page.
235 * lock_page() might sleep, the caller should not hold a spinlock.
238 get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key, int rw)
240 unsigned long address = (unsigned long)uaddr;
241 struct mm_struct *mm = current->mm;
242 struct page *page, *page_head;
246 * The futex address must be "naturally" aligned.
248 key->both.offset = address % PAGE_SIZE;
249 if (unlikely((address % sizeof(u32)) != 0))
251 address -= key->both.offset;
254 * PROCESS_PRIVATE futexes are fast.
255 * As the mm cannot disappear under us and the 'key' only needs
256 * virtual address, we dont even have to find the underlying vma.
257 * Note : We do have to check 'uaddr' is a valid user address,
258 * but access_ok() should be faster than find_vma()
261 if (unlikely(!access_ok(VERIFY_WRITE, uaddr, sizeof(u32))))
263 key->private.mm = mm;
264 key->private.address = address;
265 get_futex_key_refs(key);
270 err = get_user_pages_fast(address, 1, 1, &page);
272 * If write access is not required (eg. FUTEX_WAIT), try
273 * and get read-only access.
275 if (err == -EFAULT && rw == VERIFY_READ) {
276 err = get_user_pages_fast(address, 1, 0, &page);
284 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
286 if (unlikely(PageTail(page))) {
288 /* serialize against __split_huge_page_splitting() */
290 if (likely(__get_user_pages_fast(address, 1, 1, &page) == 1)) {
291 page_head = compound_head(page);
293 * page_head is valid pointer but we must pin
294 * it before taking the PG_lock and/or
295 * PG_compound_lock. The moment we re-enable
296 * irqs __split_huge_page_splitting() can
297 * return and the head page can be freed from
298 * under us. We can't take the PG_lock and/or
299 * PG_compound_lock on a page that could be
300 * freed from under us.
302 if (page != page_head) {
313 page_head = compound_head(page);
314 if (page != page_head) {
320 lock_page(page_head);
323 * If page_head->mapping is NULL, then it cannot be a PageAnon
324 * page; but it might be the ZERO_PAGE or in the gate area or
325 * in a special mapping (all cases which we are happy to fail);
326 * or it may have been a good file page when get_user_pages_fast
327 * found it, but truncated or holepunched or subjected to
328 * invalidate_complete_page2 before we got the page lock (also
329 * cases which we are happy to fail). And we hold a reference,
330 * so refcount care in invalidate_complete_page's remove_mapping
331 * prevents drop_caches from setting mapping to NULL beneath us.
333 * The case we do have to guard against is when memory pressure made
334 * shmem_writepage move it from filecache to swapcache beneath us:
335 * an unlikely race, but we do need to retry for page_head->mapping.
337 if (!page_head->mapping) {
338 int shmem_swizzled = PageSwapCache(page_head);
339 unlock_page(page_head);
347 * Private mappings are handled in a simple way.
349 * NOTE: When userspace waits on a MAP_SHARED mapping, even if
350 * it's a read-only handle, it's expected that futexes attach to
351 * the object not the particular process.
353 if (PageAnon(page_head)) {
355 * A RO anonymous page will never change and thus doesn't make
356 * sense for futex operations.
363 key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */
364 key->private.mm = mm;
365 key->private.address = address;
367 key->both.offset |= FUT_OFF_INODE; /* inode-based key */
368 key->shared.inode = page_head->mapping->host;
369 key->shared.pgoff = page_head->index;
372 get_futex_key_refs(key);
375 unlock_page(page_head);
380 static inline void put_futex_key(union futex_key *key)
382 drop_futex_key_refs(key);
386 * fault_in_user_writeable() - Fault in user address and verify RW access
387 * @uaddr: pointer to faulting user space address
389 * Slow path to fixup the fault we just took in the atomic write
392 * We have no generic implementation of a non-destructive write to the
393 * user address. We know that we faulted in the atomic pagefault
394 * disabled section so we can as well avoid the #PF overhead by
395 * calling get_user_pages() right away.
397 static int fault_in_user_writeable(u32 __user *uaddr)
399 struct mm_struct *mm = current->mm;
402 down_read(&mm->mmap_sem);
403 ret = fixup_user_fault(current, mm, (unsigned long)uaddr,
405 up_read(&mm->mmap_sem);
407 return ret < 0 ? ret : 0;
411 * futex_top_waiter() - Return the highest priority waiter on a futex
412 * @hb: the hash bucket the futex_q's reside in
413 * @key: the futex key (to distinguish it from other futex futex_q's)
415 * Must be called with the hb lock held.
417 static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb,
418 union futex_key *key)
420 struct futex_q *this;
422 plist_for_each_entry(this, &hb->chain, list) {
423 if (match_futex(&this->key, key))
429 static int cmpxchg_futex_value_locked(u32 *curval, u32 __user *uaddr,
430 u32 uval, u32 newval)
435 ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval);
441 static int get_futex_value_locked(u32 *dest, u32 __user *from)
446 ret = __copy_from_user_inatomic(dest, from, sizeof(u32));
449 return ret ? -EFAULT : 0;
456 static int refill_pi_state_cache(void)
458 struct futex_pi_state *pi_state;
460 if (likely(current->pi_state_cache))
463 pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL);
468 INIT_LIST_HEAD(&pi_state->list);
469 /* pi_mutex gets initialized later */
470 pi_state->owner = NULL;
471 atomic_set(&pi_state->refcount, 1);
472 pi_state->key = FUTEX_KEY_INIT;
474 current->pi_state_cache = pi_state;
479 static struct futex_pi_state * alloc_pi_state(void)
481 struct futex_pi_state *pi_state = current->pi_state_cache;
484 current->pi_state_cache = NULL;
489 static void free_pi_state(struct futex_pi_state *pi_state)
491 if (!atomic_dec_and_test(&pi_state->refcount))
495 * If pi_state->owner is NULL, the owner is most probably dying
496 * and has cleaned up the pi_state already
498 if (pi_state->owner) {
499 raw_spin_lock_irq(&pi_state->owner->pi_lock);
500 list_del_init(&pi_state->list);
501 raw_spin_unlock_irq(&pi_state->owner->pi_lock);
503 rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner);
506 if (current->pi_state_cache)
510 * pi_state->list is already empty.
511 * clear pi_state->owner.
512 * refcount is at 0 - put it back to 1.
514 pi_state->owner = NULL;
515 atomic_set(&pi_state->refcount, 1);
516 current->pi_state_cache = pi_state;
521 * Look up the task based on what TID userspace gave us.
524 static struct task_struct * futex_find_get_task(pid_t pid)
526 struct task_struct *p;
529 p = find_task_by_vpid(pid);
539 * This task is holding PI mutexes at exit time => bad.
540 * Kernel cleans up PI-state, but userspace is likely hosed.
541 * (Robust-futex cleanup is separate and might save the day for userspace.)
543 void exit_pi_state_list(struct task_struct *curr)
545 struct list_head *next, *head = &curr->pi_state_list;
546 struct futex_pi_state *pi_state;
547 struct futex_hash_bucket *hb;
548 union futex_key key = FUTEX_KEY_INIT;
550 if (!futex_cmpxchg_enabled)
553 * We are a ZOMBIE and nobody can enqueue itself on
554 * pi_state_list anymore, but we have to be careful
555 * versus waiters unqueueing themselves:
557 raw_spin_lock_irq(&curr->pi_lock);
558 while (!list_empty(head)) {
561 pi_state = list_entry(next, struct futex_pi_state, list);
563 hb = hash_futex(&key);
564 raw_spin_unlock_irq(&curr->pi_lock);
566 spin_lock(&hb->lock);
568 raw_spin_lock_irq(&curr->pi_lock);
570 * We dropped the pi-lock, so re-check whether this
571 * task still owns the PI-state:
573 if (head->next != next) {
574 spin_unlock(&hb->lock);
578 WARN_ON(pi_state->owner != curr);
579 WARN_ON(list_empty(&pi_state->list));
580 list_del_init(&pi_state->list);
581 pi_state->owner = NULL;
582 raw_spin_unlock_irq(&curr->pi_lock);
584 rt_mutex_unlock(&pi_state->pi_mutex);
586 spin_unlock(&hb->lock);
588 raw_spin_lock_irq(&curr->pi_lock);
590 raw_spin_unlock_irq(&curr->pi_lock);
594 lookup_pi_state(u32 uval, struct futex_hash_bucket *hb,
595 union futex_key *key, struct futex_pi_state **ps)
597 struct futex_pi_state *pi_state = NULL;
598 struct futex_q *this, *next;
599 struct plist_head *head;
600 struct task_struct *p;
601 pid_t pid = uval & FUTEX_TID_MASK;
605 plist_for_each_entry_safe(this, next, head, list) {
606 if (match_futex(&this->key, key)) {
608 * Another waiter already exists - bump up
609 * the refcount and return its pi_state:
611 pi_state = this->pi_state;
613 * Userspace might have messed up non-PI and PI futexes
615 if (unlikely(!pi_state))
618 WARN_ON(!atomic_read(&pi_state->refcount));
621 * When pi_state->owner is NULL then the owner died
622 * and another waiter is on the fly. pi_state->owner
623 * is fixed up by the task which acquires
624 * pi_state->rt_mutex.
626 * We do not check for pid == 0 which can happen when
627 * the owner died and robust_list_exit() cleared the
630 if (pid && pi_state->owner) {
632 * Bail out if user space manipulated the
635 if (pid != task_pid_vnr(pi_state->owner))
639 atomic_inc(&pi_state->refcount);
647 * We are the first waiter - try to look up the real owner and attach
648 * the new pi_state to it, but bail out when TID = 0
652 p = futex_find_get_task(pid);
657 * We need to look at the task state flags to figure out,
658 * whether the task is exiting. To protect against the do_exit
659 * change of the task flags, we do this protected by
662 raw_spin_lock_irq(&p->pi_lock);
663 if (unlikely(p->flags & PF_EXITING)) {
665 * The task is on the way out. When PF_EXITPIDONE is
666 * set, we know that the task has finished the
669 int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN;
671 raw_spin_unlock_irq(&p->pi_lock);
676 pi_state = alloc_pi_state();
679 * Initialize the pi_mutex in locked state and make 'p'
682 rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p);
684 /* Store the key for possible exit cleanups: */
685 pi_state->key = *key;
687 WARN_ON(!list_empty(&pi_state->list));
688 list_add(&pi_state->list, &p->pi_state_list);
690 raw_spin_unlock_irq(&p->pi_lock);
700 * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex
701 * @uaddr: the pi futex user address
702 * @hb: the pi futex hash bucket
703 * @key: the futex key associated with uaddr and hb
704 * @ps: the pi_state pointer where we store the result of the
706 * @task: the task to perform the atomic lock work for. This will
707 * be "current" except in the case of requeue pi.
708 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
712 * 1 - acquired the lock;
715 * The hb->lock and futex_key refs shall be held by the caller.
717 static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb,
718 union futex_key *key,
719 struct futex_pi_state **ps,
720 struct task_struct *task, int set_waiters)
722 int lock_taken, ret, force_take = 0;
723 u32 uval, newval, curval, vpid = task_pid_vnr(task);
726 ret = lock_taken = 0;
729 * To avoid races, we attempt to take the lock here again
730 * (by doing a 0 -> TID atomic cmpxchg), while holding all
731 * the locks. It will most likely not succeed.
735 newval |= FUTEX_WAITERS;
737 if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, 0, newval)))
743 if ((unlikely((curval & FUTEX_TID_MASK) == vpid)))
747 * Surprise - we got the lock, but we do not trust user space at all.
749 if (unlikely(!curval)) {
751 * We verify whether there is kernel state for this
752 * futex. If not, we can safely assume, that the 0 ->
753 * TID transition is correct. If state exists, we do
754 * not bother to fixup the user space state as it was
757 return futex_top_waiter(hb, key) ? -EINVAL : 1;
763 * Set the FUTEX_WAITERS flag, so the owner will know it has someone
764 * to wake at the next unlock.
766 newval = curval | FUTEX_WAITERS;
769 * Should we force take the futex? See below.
771 if (unlikely(force_take)) {
773 * Keep the OWNER_DIED and the WAITERS bit and set the
776 newval = (curval & ~FUTEX_TID_MASK) | vpid;
781 if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)))
783 if (unlikely(curval != uval))
787 * We took the lock due to forced take over.
789 if (unlikely(lock_taken))
793 * We dont have the lock. Look up the PI state (or create it if
794 * we are the first waiter):
796 ret = lookup_pi_state(uval, hb, key, ps);
802 * We failed to find an owner for this
803 * futex. So we have no pi_state to block
804 * on. This can happen in two cases:
807 * 2) A stale FUTEX_WAITERS bit
809 * Re-read the futex value.
811 if (get_futex_value_locked(&curval, uaddr))
815 * If the owner died or we have a stale
816 * WAITERS bit the owner TID in the user space
819 if (!(curval & FUTEX_TID_MASK)) {
832 * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket
833 * @q: The futex_q to unqueue
835 * The q->lock_ptr must not be NULL and must be held by the caller.
837 static void __unqueue_futex(struct futex_q *q)
839 struct futex_hash_bucket *hb;
841 if (WARN_ON_SMP(!q->lock_ptr || !spin_is_locked(q->lock_ptr))
842 || WARN_ON(plist_node_empty(&q->list)))
845 hb = container_of(q->lock_ptr, struct futex_hash_bucket, lock);
846 plist_del(&q->list, &hb->chain);
850 * The hash bucket lock must be held when this is called.
851 * Afterwards, the futex_q must not be accessed.
853 static void wake_futex(struct futex_q *q)
855 struct task_struct *p = q->task;
857 if (WARN(q->pi_state || q->rt_waiter, "refusing to wake PI futex\n"))
861 * We set q->lock_ptr = NULL _before_ we wake up the task. If
862 * a non-futex wake up happens on another CPU then the task
863 * might exit and p would dereference a non-existing task
864 * struct. Prevent this by holding a reference on p across the
871 * The waiting task can free the futex_q as soon as
872 * q->lock_ptr = NULL is written, without taking any locks. A
873 * memory barrier is required here to prevent the following
874 * store to lock_ptr from getting ahead of the plist_del.
879 wake_up_state(p, TASK_NORMAL);
883 static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this)
885 struct task_struct *new_owner;
886 struct futex_pi_state *pi_state = this->pi_state;
887 u32 uninitialized_var(curval), newval;
893 * If current does not own the pi_state then the futex is
894 * inconsistent and user space fiddled with the futex value.
896 if (pi_state->owner != current)
899 raw_spin_lock(&pi_state->pi_mutex.wait_lock);
900 new_owner = rt_mutex_next_owner(&pi_state->pi_mutex);
903 * It is possible that the next waiter (the one that brought
904 * this owner to the kernel) timed out and is no longer
905 * waiting on the lock.
908 new_owner = this->task;
911 * We pass it to the next owner. (The WAITERS bit is always
912 * kept enabled while there is PI state around. We must also
913 * preserve the owner died bit.)
915 if (!(uval & FUTEX_OWNER_DIED)) {
918 newval = FUTEX_WAITERS | task_pid_vnr(new_owner);
920 if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))
922 else if (curval != uval)
925 raw_spin_unlock(&pi_state->pi_mutex.wait_lock);
930 raw_spin_lock_irq(&pi_state->owner->pi_lock);
931 WARN_ON(list_empty(&pi_state->list));
932 list_del_init(&pi_state->list);
933 raw_spin_unlock_irq(&pi_state->owner->pi_lock);
935 raw_spin_lock_irq(&new_owner->pi_lock);
936 WARN_ON(!list_empty(&pi_state->list));
937 list_add(&pi_state->list, &new_owner->pi_state_list);
938 pi_state->owner = new_owner;
939 raw_spin_unlock_irq(&new_owner->pi_lock);
941 raw_spin_unlock(&pi_state->pi_mutex.wait_lock);
942 rt_mutex_unlock(&pi_state->pi_mutex);
947 static int unlock_futex_pi(u32 __user *uaddr, u32 uval)
949 u32 uninitialized_var(oldval);
952 * There is no waiter, so we unlock the futex. The owner died
953 * bit has not to be preserved here. We are the owner:
955 if (cmpxchg_futex_value_locked(&oldval, uaddr, uval, 0))
964 * Express the locking dependencies for lockdep:
967 double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
970 spin_lock(&hb1->lock);
972 spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING);
973 } else { /* hb1 > hb2 */
974 spin_lock(&hb2->lock);
975 spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING);
980 double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
982 spin_unlock(&hb1->lock);
984 spin_unlock(&hb2->lock);
988 * Wake up waiters matching bitset queued on this futex (uaddr).
991 futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset)
993 struct futex_hash_bucket *hb;
994 struct futex_q *this, *next;
995 struct plist_head *head;
996 union futex_key key = FUTEX_KEY_INIT;
1002 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_READ);
1003 if (unlikely(ret != 0))
1006 hb = hash_futex(&key);
1007 spin_lock(&hb->lock);
1010 plist_for_each_entry_safe(this, next, head, list) {
1011 if (match_futex (&this->key, &key)) {
1012 if (this->pi_state || this->rt_waiter) {
1017 /* Check if one of the bits is set in both bitsets */
1018 if (!(this->bitset & bitset))
1022 if (++ret >= nr_wake)
1027 spin_unlock(&hb->lock);
1028 put_futex_key(&key);
1034 * Wake up all waiters hashed on the physical page that is mapped
1035 * to this virtual address:
1038 futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2,
1039 int nr_wake, int nr_wake2, int op)
1041 union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
1042 struct futex_hash_bucket *hb1, *hb2;
1043 struct plist_head *head;
1044 struct futex_q *this, *next;
1048 ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ);
1049 if (unlikely(ret != 0))
1051 ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE);
1052 if (unlikely(ret != 0))
1055 hb1 = hash_futex(&key1);
1056 hb2 = hash_futex(&key2);
1059 double_lock_hb(hb1, hb2);
1060 op_ret = futex_atomic_op_inuser(op, uaddr2);
1061 if (unlikely(op_ret < 0)) {
1063 double_unlock_hb(hb1, hb2);
1067 * we don't get EFAULT from MMU faults if we don't have an MMU,
1068 * but we might get them from range checking
1074 if (unlikely(op_ret != -EFAULT)) {
1079 ret = fault_in_user_writeable(uaddr2);
1083 if (!(flags & FLAGS_SHARED))
1086 put_futex_key(&key2);
1087 put_futex_key(&key1);
1093 plist_for_each_entry_safe(this, next, head, list) {
1094 if (match_futex (&this->key, &key1)) {
1095 if (this->pi_state || this->rt_waiter) {
1100 if (++ret >= nr_wake)
1109 plist_for_each_entry_safe(this, next, head, list) {
1110 if (match_futex (&this->key, &key2)) {
1111 if (this->pi_state || this->rt_waiter) {
1116 if (++op_ret >= nr_wake2)
1124 double_unlock_hb(hb1, hb2);
1126 put_futex_key(&key2);
1128 put_futex_key(&key1);
1134 * requeue_futex() - Requeue a futex_q from one hb to another
1135 * @q: the futex_q to requeue
1136 * @hb1: the source hash_bucket
1137 * @hb2: the target hash_bucket
1138 * @key2: the new key for the requeued futex_q
1141 void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1,
1142 struct futex_hash_bucket *hb2, union futex_key *key2)
1146 * If key1 and key2 hash to the same bucket, no need to
1149 if (likely(&hb1->chain != &hb2->chain)) {
1150 plist_del(&q->list, &hb1->chain);
1151 plist_add(&q->list, &hb2->chain);
1152 q->lock_ptr = &hb2->lock;
1154 get_futex_key_refs(key2);
1159 * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue
1161 * @key: the key of the requeue target futex
1162 * @hb: the hash_bucket of the requeue target futex
1164 * During futex_requeue, with requeue_pi=1, it is possible to acquire the
1165 * target futex if it is uncontended or via a lock steal. Set the futex_q key
1166 * to the requeue target futex so the waiter can detect the wakeup on the right
1167 * futex, but remove it from the hb and NULL the rt_waiter so it can detect
1168 * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock
1169 * to protect access to the pi_state to fixup the owner later. Must be called
1170 * with both q->lock_ptr and hb->lock held.
1173 void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key,
1174 struct futex_hash_bucket *hb)
1176 get_futex_key_refs(key);
1181 WARN_ON(!q->rt_waiter);
1182 q->rt_waiter = NULL;
1184 q->lock_ptr = &hb->lock;
1186 wake_up_state(q->task, TASK_NORMAL);
1190 * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter
1191 * @pifutex: the user address of the to futex
1192 * @hb1: the from futex hash bucket, must be locked by the caller
1193 * @hb2: the to futex hash bucket, must be locked by the caller
1194 * @key1: the from futex key
1195 * @key2: the to futex key
1196 * @ps: address to store the pi_state pointer
1197 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
1199 * Try and get the lock on behalf of the top waiter if we can do it atomically.
1200 * Wake the top waiter if we succeed. If the caller specified set_waiters,
1201 * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit.
1202 * hb1 and hb2 must be held by the caller.
1205 * 0 - failed to acquire the lock atomically;
1206 * 1 - acquired the lock;
1209 static int futex_proxy_trylock_atomic(u32 __user *pifutex,
1210 struct futex_hash_bucket *hb1,
1211 struct futex_hash_bucket *hb2,
1212 union futex_key *key1, union futex_key *key2,
1213 struct futex_pi_state **ps, int set_waiters)
1215 struct futex_q *top_waiter = NULL;
1219 if (get_futex_value_locked(&curval, pifutex))
1223 * Find the top_waiter and determine if there are additional waiters.
1224 * If the caller intends to requeue more than 1 waiter to pifutex,
1225 * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now,
1226 * as we have means to handle the possible fault. If not, don't set
1227 * the bit unecessarily as it will force the subsequent unlock to enter
1230 top_waiter = futex_top_waiter(hb1, key1);
1232 /* There are no waiters, nothing for us to do. */
1236 /* Ensure we requeue to the expected futex. */
1237 if (!match_futex(top_waiter->requeue_pi_key, key2))
1241 * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in
1242 * the contended case or if set_waiters is 1. The pi_state is returned
1243 * in ps in contended cases.
1245 ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task,
1248 requeue_pi_wake_futex(top_waiter, key2, hb2);
1254 * futex_requeue() - Requeue waiters from uaddr1 to uaddr2
1255 * @uaddr1: source futex user address
1256 * @flags: futex flags (FLAGS_SHARED, etc.)
1257 * @uaddr2: target futex user address
1258 * @nr_wake: number of waiters to wake (must be 1 for requeue_pi)
1259 * @nr_requeue: number of waiters to requeue (0-INT_MAX)
1260 * @cmpval: @uaddr1 expected value (or %NULL)
1261 * @requeue_pi: if we are attempting to requeue from a non-pi futex to a
1262 * pi futex (pi to pi requeue is not supported)
1264 * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire
1265 * uaddr2 atomically on behalf of the top waiter.
1268 * >=0 - on success, the number of tasks requeued or woken;
1271 static int futex_requeue(u32 __user *uaddr1, unsigned int flags,
1272 u32 __user *uaddr2, int nr_wake, int nr_requeue,
1273 u32 *cmpval, int requeue_pi)
1275 union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
1276 int drop_count = 0, task_count = 0, ret;
1277 struct futex_pi_state *pi_state = NULL;
1278 struct futex_hash_bucket *hb1, *hb2;
1279 struct plist_head *head1;
1280 struct futex_q *this, *next;
1285 * Requeue PI only works on two distinct uaddrs. This
1286 * check is only valid for private futexes. See below.
1288 if (uaddr1 == uaddr2)
1292 * requeue_pi requires a pi_state, try to allocate it now
1293 * without any locks in case it fails.
1295 if (refill_pi_state_cache())
1298 * requeue_pi must wake as many tasks as it can, up to nr_wake
1299 * + nr_requeue, since it acquires the rt_mutex prior to
1300 * returning to userspace, so as to not leave the rt_mutex with
1301 * waiters and no owner. However, second and third wake-ups
1302 * cannot be predicted as they involve race conditions with the
1303 * first wake and a fault while looking up the pi_state. Both
1304 * pthread_cond_signal() and pthread_cond_broadcast() should
1312 if (pi_state != NULL) {
1314 * We will have to lookup the pi_state again, so free this one
1315 * to keep the accounting correct.
1317 free_pi_state(pi_state);
1321 ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ);
1322 if (unlikely(ret != 0))
1324 ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2,
1325 requeue_pi ? VERIFY_WRITE : VERIFY_READ);
1326 if (unlikely(ret != 0))
1330 * The check above which compares uaddrs is not sufficient for
1331 * shared futexes. We need to compare the keys:
1333 if (requeue_pi && match_futex(&key1, &key2)) {
1338 hb1 = hash_futex(&key1);
1339 hb2 = hash_futex(&key2);
1342 double_lock_hb(hb1, hb2);
1344 if (likely(cmpval != NULL)) {
1347 ret = get_futex_value_locked(&curval, uaddr1);
1349 if (unlikely(ret)) {
1350 double_unlock_hb(hb1, hb2);
1352 ret = get_user(curval, uaddr1);
1356 if (!(flags & FLAGS_SHARED))
1359 put_futex_key(&key2);
1360 put_futex_key(&key1);
1363 if (curval != *cmpval) {
1369 if (requeue_pi && (task_count - nr_wake < nr_requeue)) {
1371 * Attempt to acquire uaddr2 and wake the top waiter. If we
1372 * intend to requeue waiters, force setting the FUTEX_WAITERS
1373 * bit. We force this here where we are able to easily handle
1374 * faults rather in the requeue loop below.
1376 ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1,
1377 &key2, &pi_state, nr_requeue);
1380 * At this point the top_waiter has either taken uaddr2 or is
1381 * waiting on it. If the former, then the pi_state will not
1382 * exist yet, look it up one more time to ensure we have a
1389 ret = get_futex_value_locked(&curval2, uaddr2);
1391 ret = lookup_pi_state(curval2, hb2, &key2,
1399 double_unlock_hb(hb1, hb2);
1400 put_futex_key(&key2);
1401 put_futex_key(&key1);
1402 ret = fault_in_user_writeable(uaddr2);
1407 /* The owner was exiting, try again. */
1408 double_unlock_hb(hb1, hb2);
1409 put_futex_key(&key2);
1410 put_futex_key(&key1);
1418 head1 = &hb1->chain;
1419 plist_for_each_entry_safe(this, next, head1, list) {
1420 if (task_count - nr_wake >= nr_requeue)
1423 if (!match_futex(&this->key, &key1))
1427 * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always
1428 * be paired with each other and no other futex ops.
1430 * We should never be requeueing a futex_q with a pi_state,
1431 * which is awaiting a futex_unlock_pi().
1433 if ((requeue_pi && !this->rt_waiter) ||
1434 (!requeue_pi && this->rt_waiter) ||
1441 * Wake nr_wake waiters. For requeue_pi, if we acquired the
1442 * lock, we already woke the top_waiter. If not, it will be
1443 * woken by futex_unlock_pi().
1445 if (++task_count <= nr_wake && !requeue_pi) {
1450 /* Ensure we requeue to the expected futex for requeue_pi. */
1451 if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) {
1457 * Requeue nr_requeue waiters and possibly one more in the case
1458 * of requeue_pi if we couldn't acquire the lock atomically.
1461 /* Prepare the waiter to take the rt_mutex. */
1462 atomic_inc(&pi_state->refcount);
1463 this->pi_state = pi_state;
1464 ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex,
1468 /* We got the lock. */
1469 requeue_pi_wake_futex(this, &key2, hb2);
1474 this->pi_state = NULL;
1475 free_pi_state(pi_state);
1479 requeue_futex(this, hb1, hb2, &key2);
1484 double_unlock_hb(hb1, hb2);
1487 * drop_futex_key_refs() must be called outside the spinlocks. During
1488 * the requeue we moved futex_q's from the hash bucket at key1 to the
1489 * one at key2 and updated their key pointer. We no longer need to
1490 * hold the references to key1.
1492 while (--drop_count >= 0)
1493 drop_futex_key_refs(&key1);
1496 put_futex_key(&key2);
1498 put_futex_key(&key1);
1500 if (pi_state != NULL)
1501 free_pi_state(pi_state);
1502 return ret ? ret : task_count;
1505 /* The key must be already stored in q->key. */
1506 static inline struct futex_hash_bucket *queue_lock(struct futex_q *q)
1507 __acquires(&hb->lock)
1509 struct futex_hash_bucket *hb;
1511 hb = hash_futex(&q->key);
1512 q->lock_ptr = &hb->lock;
1514 spin_lock(&hb->lock);
1519 queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb)
1520 __releases(&hb->lock)
1522 spin_unlock(&hb->lock);
1526 * queue_me() - Enqueue the futex_q on the futex_hash_bucket
1527 * @q: The futex_q to enqueue
1528 * @hb: The destination hash bucket
1530 * The hb->lock must be held by the caller, and is released here. A call to
1531 * queue_me() is typically paired with exactly one call to unqueue_me(). The
1532 * exceptions involve the PI related operations, which may use unqueue_me_pi()
1533 * or nothing if the unqueue is done as part of the wake process and the unqueue
1534 * state is implicit in the state of woken task (see futex_wait_requeue_pi() for
1537 static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb)
1538 __releases(&hb->lock)
1543 * The priority used to register this element is
1544 * - either the real thread-priority for the real-time threads
1545 * (i.e. threads with a priority lower than MAX_RT_PRIO)
1546 * - or MAX_RT_PRIO for non-RT threads.
1547 * Thus, all RT-threads are woken first in priority order, and
1548 * the others are woken last, in FIFO order.
1550 prio = min(current->normal_prio, MAX_RT_PRIO);
1552 plist_node_init(&q->list, prio);
1553 plist_add(&q->list, &hb->chain);
1555 spin_unlock(&hb->lock);
1559 * unqueue_me() - Remove the futex_q from its futex_hash_bucket
1560 * @q: The futex_q to unqueue
1562 * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must
1563 * be paired with exactly one earlier call to queue_me().
1566 * 1 - if the futex_q was still queued (and we removed unqueued it);
1567 * 0 - if the futex_q was already removed by the waking thread
1569 static int unqueue_me(struct futex_q *q)
1571 spinlock_t *lock_ptr;
1574 /* In the common case we don't take the spinlock, which is nice. */
1576 lock_ptr = q->lock_ptr;
1578 if (lock_ptr != NULL) {
1579 spin_lock(lock_ptr);
1581 * q->lock_ptr can change between reading it and
1582 * spin_lock(), causing us to take the wrong lock. This
1583 * corrects the race condition.
1585 * Reasoning goes like this: if we have the wrong lock,
1586 * q->lock_ptr must have changed (maybe several times)
1587 * between reading it and the spin_lock(). It can
1588 * change again after the spin_lock() but only if it was
1589 * already changed before the spin_lock(). It cannot,
1590 * however, change back to the original value. Therefore
1591 * we can detect whether we acquired the correct lock.
1593 if (unlikely(lock_ptr != q->lock_ptr)) {
1594 spin_unlock(lock_ptr);
1599 BUG_ON(q->pi_state);
1601 spin_unlock(lock_ptr);
1605 drop_futex_key_refs(&q->key);
1610 * PI futexes can not be requeued and must remove themself from the
1611 * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry
1614 static void unqueue_me_pi(struct futex_q *q)
1615 __releases(q->lock_ptr)
1619 BUG_ON(!q->pi_state);
1620 free_pi_state(q->pi_state);
1623 spin_unlock(q->lock_ptr);
1627 * Fixup the pi_state owner with the new owner.
1629 * Must be called with hash bucket lock held and mm->sem held for non
1632 static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q,
1633 struct task_struct *newowner)
1635 u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS;
1636 struct futex_pi_state *pi_state = q->pi_state;
1637 struct task_struct *oldowner = pi_state->owner;
1638 u32 uval, uninitialized_var(curval), newval;
1642 if (!pi_state->owner)
1643 newtid |= FUTEX_OWNER_DIED;
1646 * We are here either because we stole the rtmutex from the
1647 * previous highest priority waiter or we are the highest priority
1648 * waiter but failed to get the rtmutex the first time.
1649 * We have to replace the newowner TID in the user space variable.
1650 * This must be atomic as we have to preserve the owner died bit here.
1652 * Note: We write the user space value _before_ changing the pi_state
1653 * because we can fault here. Imagine swapped out pages or a fork
1654 * that marked all the anonymous memory readonly for cow.
1656 * Modifying pi_state _before_ the user space value would
1657 * leave the pi_state in an inconsistent state when we fault
1658 * here, because we need to drop the hash bucket lock to
1659 * handle the fault. This might be observed in the PID check
1660 * in lookup_pi_state.
1663 if (get_futex_value_locked(&uval, uaddr))
1667 newval = (uval & FUTEX_OWNER_DIED) | newtid;
1669 if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))
1677 * We fixed up user space. Now we need to fix the pi_state
1680 if (pi_state->owner != NULL) {
1681 raw_spin_lock_irq(&pi_state->owner->pi_lock);
1682 WARN_ON(list_empty(&pi_state->list));
1683 list_del_init(&pi_state->list);
1684 raw_spin_unlock_irq(&pi_state->owner->pi_lock);
1687 pi_state->owner = newowner;
1689 raw_spin_lock_irq(&newowner->pi_lock);
1690 WARN_ON(!list_empty(&pi_state->list));
1691 list_add(&pi_state->list, &newowner->pi_state_list);
1692 raw_spin_unlock_irq(&newowner->pi_lock);
1696 * To handle the page fault we need to drop the hash bucket
1697 * lock here. That gives the other task (either the highest priority
1698 * waiter itself or the task which stole the rtmutex) the
1699 * chance to try the fixup of the pi_state. So once we are
1700 * back from handling the fault we need to check the pi_state
1701 * after reacquiring the hash bucket lock and before trying to
1702 * do another fixup. When the fixup has been done already we
1706 spin_unlock(q->lock_ptr);
1708 ret = fault_in_user_writeable(uaddr);
1710 spin_lock(q->lock_ptr);
1713 * Check if someone else fixed it for us:
1715 if (pi_state->owner != oldowner)
1724 static long futex_wait_restart(struct restart_block *restart);
1727 * fixup_owner() - Post lock pi_state and corner case management
1728 * @uaddr: user address of the futex
1729 * @q: futex_q (contains pi_state and access to the rt_mutex)
1730 * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0)
1732 * After attempting to lock an rt_mutex, this function is called to cleanup
1733 * the pi_state owner as well as handle race conditions that may allow us to
1734 * acquire the lock. Must be called with the hb lock held.
1737 * 1 - success, lock taken;
1738 * 0 - success, lock not taken;
1739 * <0 - on error (-EFAULT)
1741 static int fixup_owner(u32 __user *uaddr, struct futex_q *q, int locked)
1743 struct task_struct *owner;
1748 * Got the lock. We might not be the anticipated owner if we
1749 * did a lock-steal - fix up the PI-state in that case:
1751 if (q->pi_state->owner != current)
1752 ret = fixup_pi_state_owner(uaddr, q, current);
1757 * Catch the rare case, where the lock was released when we were on the
1758 * way back before we locked the hash bucket.
1760 if (q->pi_state->owner == current) {
1762 * Try to get the rt_mutex now. This might fail as some other
1763 * task acquired the rt_mutex after we removed ourself from the
1764 * rt_mutex waiters list.
1766 if (rt_mutex_trylock(&q->pi_state->pi_mutex)) {
1772 * pi_state is incorrect, some other task did a lock steal and
1773 * we returned due to timeout or signal without taking the
1774 * rt_mutex. Too late.
1776 raw_spin_lock(&q->pi_state->pi_mutex.wait_lock);
1777 owner = rt_mutex_owner(&q->pi_state->pi_mutex);
1779 owner = rt_mutex_next_owner(&q->pi_state->pi_mutex);
1780 raw_spin_unlock(&q->pi_state->pi_mutex.wait_lock);
1781 ret = fixup_pi_state_owner(uaddr, q, owner);
1786 * Paranoia check. If we did not take the lock, then we should not be
1787 * the owner of the rt_mutex.
1789 if (rt_mutex_owner(&q->pi_state->pi_mutex) == current)
1790 printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p "
1791 "pi-state %p\n", ret,
1792 q->pi_state->pi_mutex.owner,
1793 q->pi_state->owner);
1796 return ret ? ret : locked;
1800 * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal
1801 * @hb: the futex hash bucket, must be locked by the caller
1802 * @q: the futex_q to queue up on
1803 * @timeout: the prepared hrtimer_sleeper, or null for no timeout
1805 static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q,
1806 struct hrtimer_sleeper *timeout)
1809 * The task state is guaranteed to be set before another task can
1810 * wake it. set_current_state() is implemented using set_mb() and
1811 * queue_me() calls spin_unlock() upon completion, both serializing
1812 * access to the hash list and forcing another memory barrier.
1814 set_current_state(TASK_INTERRUPTIBLE);
1819 hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS);
1820 if (!hrtimer_active(&timeout->timer))
1821 timeout->task = NULL;
1825 * If we have been removed from the hash list, then another task
1826 * has tried to wake us, and we can skip the call to schedule().
1828 if (likely(!plist_node_empty(&q->list))) {
1830 * If the timer has already expired, current will already be
1831 * flagged for rescheduling. Only call schedule if there
1832 * is no timeout, or if it has yet to expire.
1834 if (!timeout || timeout->task)
1835 freezable_schedule();
1837 __set_current_state(TASK_RUNNING);
1841 * futex_wait_setup() - Prepare to wait on a futex
1842 * @uaddr: the futex userspace address
1843 * @val: the expected value
1844 * @flags: futex flags (FLAGS_SHARED, etc.)
1845 * @q: the associated futex_q
1846 * @hb: storage for hash_bucket pointer to be returned to caller
1848 * Setup the futex_q and locate the hash_bucket. Get the futex value and
1849 * compare it with the expected value. Handle atomic faults internally.
1850 * Return with the hb lock held and a q.key reference on success, and unlocked
1851 * with no q.key reference on failure.
1854 * 0 - uaddr contains val and hb has been locked;
1855 * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlocked
1857 static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags,
1858 struct futex_q *q, struct futex_hash_bucket **hb)
1864 * Access the page AFTER the hash-bucket is locked.
1865 * Order is important:
1867 * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val);
1868 * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); }
1870 * The basic logical guarantee of a futex is that it blocks ONLY
1871 * if cond(var) is known to be true at the time of blocking, for
1872 * any cond. If we locked the hash-bucket after testing *uaddr, that
1873 * would open a race condition where we could block indefinitely with
1874 * cond(var) false, which would violate the guarantee.
1876 * On the other hand, we insert q and release the hash-bucket only
1877 * after testing *uaddr. This guarantees that futex_wait() will NOT
1878 * absorb a wakeup if *uaddr does not match the desired values
1879 * while the syscall executes.
1882 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q->key, VERIFY_READ);
1883 if (unlikely(ret != 0))
1887 *hb = queue_lock(q);
1889 ret = get_futex_value_locked(&uval, uaddr);
1892 queue_unlock(q, *hb);
1894 ret = get_user(uval, uaddr);
1898 if (!(flags & FLAGS_SHARED))
1901 put_futex_key(&q->key);
1906 queue_unlock(q, *hb);
1912 put_futex_key(&q->key);
1916 static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val,
1917 ktime_t *abs_time, u32 bitset)
1919 struct hrtimer_sleeper timeout, *to = NULL;
1920 struct restart_block *restart;
1921 struct futex_hash_bucket *hb;
1922 struct futex_q q = futex_q_init;
1932 hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ?
1933 CLOCK_REALTIME : CLOCK_MONOTONIC,
1935 hrtimer_init_sleeper(to, current);
1936 hrtimer_set_expires_range_ns(&to->timer, *abs_time,
1937 current->timer_slack_ns);
1942 * Prepare to wait on uaddr. On success, holds hb lock and increments
1945 ret = futex_wait_setup(uaddr, val, flags, &q, &hb);
1949 /* queue_me and wait for wakeup, timeout, or a signal. */
1950 futex_wait_queue_me(hb, &q, to);
1952 /* If we were woken (and unqueued), we succeeded, whatever. */
1954 /* unqueue_me() drops q.key ref */
1955 if (!unqueue_me(&q))
1958 if (to && !to->task)
1962 * We expect signal_pending(current), but we might be the
1963 * victim of a spurious wakeup as well.
1965 if (!signal_pending(current))
1972 restart = ¤t_thread_info()->restart_block;
1973 restart->fn = futex_wait_restart;
1974 restart->futex.uaddr = uaddr;
1975 restart->futex.val = val;
1976 restart->futex.time = abs_time->tv64;
1977 restart->futex.bitset = bitset;
1978 restart->futex.flags = flags | FLAGS_HAS_TIMEOUT;
1980 ret = -ERESTART_RESTARTBLOCK;
1984 hrtimer_cancel(&to->timer);
1985 destroy_hrtimer_on_stack(&to->timer);
1991 static long futex_wait_restart(struct restart_block *restart)
1993 u32 __user *uaddr = restart->futex.uaddr;
1994 ktime_t t, *tp = NULL;
1996 if (restart->futex.flags & FLAGS_HAS_TIMEOUT) {
1997 t.tv64 = restart->futex.time;
2000 restart->fn = do_no_restart_syscall;
2002 return (long)futex_wait(uaddr, restart->futex.flags,
2003 restart->futex.val, tp, restart->futex.bitset);
2008 * Userspace tried a 0 -> TID atomic transition of the futex value
2009 * and failed. The kernel side here does the whole locking operation:
2010 * if there are waiters then it will block, it does PI, etc. (Due to
2011 * races the kernel might see a 0 value of the futex too.)
2013 static int futex_lock_pi(u32 __user *uaddr, unsigned int flags, int detect,
2014 ktime_t *time, int trylock)
2016 struct hrtimer_sleeper timeout, *to = NULL;
2017 struct futex_hash_bucket *hb;
2018 struct futex_q q = futex_q_init;
2021 if (refill_pi_state_cache())
2026 hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME,
2028 hrtimer_init_sleeper(to, current);
2029 hrtimer_set_expires(&to->timer, *time);
2033 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q.key, VERIFY_WRITE);
2034 if (unlikely(ret != 0))
2038 hb = queue_lock(&q);
2040 ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, 0);
2041 if (unlikely(ret)) {
2044 /* We got the lock. */
2046 goto out_unlock_put_key;
2051 * Task is exiting and we just wait for the
2054 queue_unlock(&q, hb);
2055 put_futex_key(&q.key);
2059 goto out_unlock_put_key;
2064 * Only actually queue now that the atomic ops are done:
2068 WARN_ON(!q.pi_state);
2070 * Block on the PI mutex:
2073 ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1);
2075 ret = rt_mutex_trylock(&q.pi_state->pi_mutex);
2076 /* Fixup the trylock return value: */
2077 ret = ret ? 0 : -EWOULDBLOCK;
2080 spin_lock(q.lock_ptr);
2082 * Fixup the pi_state owner and possibly acquire the lock if we
2085 res = fixup_owner(uaddr, &q, !ret);
2087 * If fixup_owner() returned an error, proprogate that. If it acquired
2088 * the lock, clear our -ETIMEDOUT or -EINTR.
2091 ret = (res < 0) ? res : 0;
2094 * If fixup_owner() faulted and was unable to handle the fault, unlock
2095 * it and return the fault to userspace.
2097 if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current))
2098 rt_mutex_unlock(&q.pi_state->pi_mutex);
2100 /* Unqueue and drop the lock */
2106 queue_unlock(&q, hb);
2109 put_futex_key(&q.key);
2112 destroy_hrtimer_on_stack(&to->timer);
2113 return ret != -EINTR ? ret : -ERESTARTNOINTR;
2116 queue_unlock(&q, hb);
2118 ret = fault_in_user_writeable(uaddr);
2122 if (!(flags & FLAGS_SHARED))
2125 put_futex_key(&q.key);
2130 * Userspace attempted a TID -> 0 atomic transition, and failed.
2131 * This is the in-kernel slowpath: we look up the PI state (if any),
2132 * and do the rt-mutex unlock.
2134 static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags)
2136 struct futex_hash_bucket *hb;
2137 struct futex_q *this, *next;
2138 struct plist_head *head;
2139 union futex_key key = FUTEX_KEY_INIT;
2140 u32 uval, vpid = task_pid_vnr(current);
2144 if (get_user(uval, uaddr))
2147 * We release only a lock we actually own:
2149 if ((uval & FUTEX_TID_MASK) != vpid)
2152 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_WRITE);
2153 if (unlikely(ret != 0))
2156 hb = hash_futex(&key);
2157 spin_lock(&hb->lock);
2160 * To avoid races, try to do the TID -> 0 atomic transition
2161 * again. If it succeeds then we can return without waking
2164 if (!(uval & FUTEX_OWNER_DIED) &&
2165 cmpxchg_futex_value_locked(&uval, uaddr, vpid, 0))
2168 * Rare case: we managed to release the lock atomically,
2169 * no need to wake anyone else up:
2171 if (unlikely(uval == vpid))
2175 * Ok, other tasks may need to be woken up - check waiters
2176 * and do the wakeup if necessary:
2180 plist_for_each_entry_safe(this, next, head, list) {
2181 if (!match_futex (&this->key, &key))
2183 ret = wake_futex_pi(uaddr, uval, this);
2185 * The atomic access to the futex value
2186 * generated a pagefault, so retry the
2187 * user-access and the wakeup:
2194 * No waiters - kernel unlocks the futex:
2196 if (!(uval & FUTEX_OWNER_DIED)) {
2197 ret = unlock_futex_pi(uaddr, uval);
2203 spin_unlock(&hb->lock);
2204 put_futex_key(&key);
2210 spin_unlock(&hb->lock);
2211 put_futex_key(&key);
2213 ret = fault_in_user_writeable(uaddr);
2221 * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex
2222 * @hb: the hash_bucket futex_q was original enqueued on
2223 * @q: the futex_q woken while waiting to be requeued
2224 * @key2: the futex_key of the requeue target futex
2225 * @timeout: the timeout associated with the wait (NULL if none)
2227 * Detect if the task was woken on the initial futex as opposed to the requeue
2228 * target futex. If so, determine if it was a timeout or a signal that caused
2229 * the wakeup and return the appropriate error code to the caller. Must be
2230 * called with the hb lock held.
2233 * 0 = no early wakeup detected;
2234 * <0 = -ETIMEDOUT or -ERESTARTNOINTR
2237 int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb,
2238 struct futex_q *q, union futex_key *key2,
2239 struct hrtimer_sleeper *timeout)
2244 * With the hb lock held, we avoid races while we process the wakeup.
2245 * We only need to hold hb (and not hb2) to ensure atomicity as the
2246 * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb.
2247 * It can't be requeued from uaddr2 to something else since we don't
2248 * support a PI aware source futex for requeue.
2250 if (!match_futex(&q->key, key2)) {
2251 WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr));
2253 * We were woken prior to requeue by a timeout or a signal.
2254 * Unqueue the futex_q and determine which it was.
2256 plist_del(&q->list, &hb->chain);
2258 /* Handle spurious wakeups gracefully */
2260 if (timeout && !timeout->task)
2262 else if (signal_pending(current))
2263 ret = -ERESTARTNOINTR;
2269 * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2
2270 * @uaddr: the futex we initially wait on (non-pi)
2271 * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be
2272 * the same type, no requeueing from private to shared, etc.
2273 * @val: the expected value of uaddr
2274 * @abs_time: absolute timeout
2275 * @bitset: 32 bit wakeup bitset set by userspace, defaults to all
2276 * @uaddr2: the pi futex we will take prior to returning to user-space
2278 * The caller will wait on uaddr and will be requeued by futex_requeue() to
2279 * uaddr2 which must be PI aware and unique from uaddr. Normal wakeup will wake
2280 * on uaddr2 and complete the acquisition of the rt_mutex prior to returning to
2281 * userspace. This ensures the rt_mutex maintains an owner when it has waiters;
2282 * without one, the pi logic would not know which task to boost/deboost, if
2283 * there was a need to.
2285 * We call schedule in futex_wait_queue_me() when we enqueue and return there
2286 * via the following--
2287 * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue()
2288 * 2) wakeup on uaddr2 after a requeue
2292 * If 3, cleanup and return -ERESTARTNOINTR.
2294 * If 2, we may then block on trying to take the rt_mutex and return via:
2295 * 5) successful lock
2298 * 8) other lock acquisition failure
2300 * If 6, return -EWOULDBLOCK (restarting the syscall would do the same).
2302 * If 4 or 7, we cleanup and return with -ETIMEDOUT.
2308 static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
2309 u32 val, ktime_t *abs_time, u32 bitset,
2312 struct hrtimer_sleeper timeout, *to = NULL;
2313 struct rt_mutex_waiter rt_waiter;
2314 struct rt_mutex *pi_mutex = NULL;
2315 struct futex_hash_bucket *hb;
2316 union futex_key key2 = FUTEX_KEY_INIT;
2317 struct futex_q q = futex_q_init;
2320 if (uaddr == uaddr2)
2328 hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ?
2329 CLOCK_REALTIME : CLOCK_MONOTONIC,
2331 hrtimer_init_sleeper(to, current);
2332 hrtimer_set_expires_range_ns(&to->timer, *abs_time,
2333 current->timer_slack_ns);
2337 * The waiter is allocated on our stack, manipulated by the requeue
2338 * code while we sleep on uaddr.
2340 debug_rt_mutex_init_waiter(&rt_waiter);
2341 rt_waiter.task = NULL;
2343 ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE);
2344 if (unlikely(ret != 0))
2348 q.rt_waiter = &rt_waiter;
2349 q.requeue_pi_key = &key2;
2352 * Prepare to wait on uaddr. On success, increments q.key (key1) ref
2355 ret = futex_wait_setup(uaddr, val, flags, &q, &hb);
2360 * The check above which compares uaddrs is not sufficient for
2361 * shared futexes. We need to compare the keys:
2363 if (match_futex(&q.key, &key2)) {
2368 /* Queue the futex_q, drop the hb lock, wait for wakeup. */
2369 futex_wait_queue_me(hb, &q, to);
2371 spin_lock(&hb->lock);
2372 ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to);
2373 spin_unlock(&hb->lock);
2378 * In order for us to be here, we know our q.key == key2, and since
2379 * we took the hb->lock above, we also know that futex_requeue() has
2380 * completed and we no longer have to concern ourselves with a wakeup
2381 * race with the atomic proxy lock acquisition by the requeue code. The
2382 * futex_requeue dropped our key1 reference and incremented our key2
2386 /* Check if the requeue code acquired the second futex for us. */
2389 * Got the lock. We might not be the anticipated owner if we
2390 * did a lock-steal - fix up the PI-state in that case.
2392 if (q.pi_state && (q.pi_state->owner != current)) {
2393 spin_lock(q.lock_ptr);
2394 ret = fixup_pi_state_owner(uaddr2, &q, current);
2395 spin_unlock(q.lock_ptr);
2399 * We have been woken up by futex_unlock_pi(), a timeout, or a
2400 * signal. futex_unlock_pi() will not destroy the lock_ptr nor
2403 WARN_ON(!q.pi_state);
2404 pi_mutex = &q.pi_state->pi_mutex;
2405 ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter, 1);
2406 debug_rt_mutex_free_waiter(&rt_waiter);
2408 spin_lock(q.lock_ptr);
2410 * Fixup the pi_state owner and possibly acquire the lock if we
2413 res = fixup_owner(uaddr2, &q, !ret);
2415 * If fixup_owner() returned an error, proprogate that. If it
2416 * acquired the lock, clear -ETIMEDOUT or -EINTR.
2419 ret = (res < 0) ? res : 0;
2421 /* Unqueue and drop the lock. */
2426 * If fixup_pi_state_owner() faulted and was unable to handle the
2427 * fault, unlock the rt_mutex and return the fault to userspace.
2429 if (ret == -EFAULT) {
2430 if (pi_mutex && rt_mutex_owner(pi_mutex) == current)
2431 rt_mutex_unlock(pi_mutex);
2432 } else if (ret == -EINTR) {
2434 * We've already been requeued, but cannot restart by calling
2435 * futex_lock_pi() directly. We could restart this syscall, but
2436 * it would detect that the user space "val" changed and return
2437 * -EWOULDBLOCK. Save the overhead of the restart and return
2438 * -EWOULDBLOCK directly.
2444 put_futex_key(&q.key);
2446 put_futex_key(&key2);
2450 hrtimer_cancel(&to->timer);
2451 destroy_hrtimer_on_stack(&to->timer);
2457 * Support for robust futexes: the kernel cleans up held futexes at
2460 * Implementation: user-space maintains a per-thread list of locks it
2461 * is holding. Upon do_exit(), the kernel carefully walks this list,
2462 * and marks all locks that are owned by this thread with the
2463 * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is
2464 * always manipulated with the lock held, so the list is private and
2465 * per-thread. Userspace also maintains a per-thread 'list_op_pending'
2466 * field, to allow the kernel to clean up if the thread dies after
2467 * acquiring the lock, but just before it could have added itself to
2468 * the list. There can only be one such pending lock.
2472 * sys_set_robust_list() - Set the robust-futex list head of a task
2473 * @head: pointer to the list-head
2474 * @len: length of the list-head, as userspace expects
2476 SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head,
2479 if (!futex_cmpxchg_enabled)
2482 * The kernel knows only one size for now:
2484 if (unlikely(len != sizeof(*head)))
2487 current->robust_list = head;
2493 * sys_get_robust_list() - Get the robust-futex list head of a task
2494 * @pid: pid of the process [zero for current task]
2495 * @head_ptr: pointer to a list-head pointer, the kernel fills it in
2496 * @len_ptr: pointer to a length field, the kernel fills in the header size
2498 SYSCALL_DEFINE3(get_robust_list, int, pid,
2499 struct robust_list_head __user * __user *, head_ptr,
2500 size_t __user *, len_ptr)
2502 struct robust_list_head __user *head;
2504 struct task_struct *p;
2506 if (!futex_cmpxchg_enabled)
2515 p = find_task_by_vpid(pid);
2521 if (!ptrace_may_access(p, PTRACE_MODE_READ))
2524 head = p->robust_list;
2527 if (put_user(sizeof(*head), len_ptr))
2529 return put_user(head, head_ptr);
2538 * Process a futex-list entry, check whether it's owned by the
2539 * dying task, and do notification if so:
2541 int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi)
2543 u32 uval, uninitialized_var(nval), mval;
2546 if (get_user(uval, uaddr))
2549 if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) {
2551 * Ok, this dying thread is truly holding a futex
2552 * of interest. Set the OWNER_DIED bit atomically
2553 * via cmpxchg, and if the value had FUTEX_WAITERS
2554 * set, wake up a waiter (if any). (We have to do a
2555 * futex_wake() even if OWNER_DIED is already set -
2556 * to handle the rare but possible case of recursive
2557 * thread-death.) The rest of the cleanup is done in
2560 mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED;
2562 * We are not holding a lock here, but we want to have
2563 * the pagefault_disable/enable() protection because
2564 * we want to handle the fault gracefully. If the
2565 * access fails we try to fault in the futex with R/W
2566 * verification via get_user_pages. get_user() above
2567 * does not guarantee R/W access. If that fails we
2568 * give up and leave the futex locked.
2570 if (cmpxchg_futex_value_locked(&nval, uaddr, uval, mval)) {
2571 if (fault_in_user_writeable(uaddr))
2579 * Wake robust non-PI futexes here. The wakeup of
2580 * PI futexes happens in exit_pi_state():
2582 if (!pi && (uval & FUTEX_WAITERS))
2583 futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY);
2589 * Fetch a robust-list pointer. Bit 0 signals PI futexes:
2591 static inline int fetch_robust_entry(struct robust_list __user **entry,
2592 struct robust_list __user * __user *head,
2595 unsigned long uentry;
2597 if (get_user(uentry, (unsigned long __user *)head))
2600 *entry = (void __user *)(uentry & ~1UL);
2607 * Walk curr->robust_list (very carefully, it's a userspace list!)
2608 * and mark any locks found there dead, and notify any waiters.
2610 * We silently return on any sign of list-walking problem.
2612 void exit_robust_list(struct task_struct *curr)
2614 struct robust_list_head __user *head = curr->robust_list;
2615 struct robust_list __user *entry, *next_entry, *pending;
2616 unsigned int limit = ROBUST_LIST_LIMIT, pi, pip;
2617 unsigned int uninitialized_var(next_pi);
2618 unsigned long futex_offset;
2621 if (!futex_cmpxchg_enabled)
2625 * Fetch the list head (which was registered earlier, via
2626 * sys_set_robust_list()):
2628 if (fetch_robust_entry(&entry, &head->list.next, &pi))
2631 * Fetch the relative futex offset:
2633 if (get_user(futex_offset, &head->futex_offset))
2636 * Fetch any possibly pending lock-add first, and handle it
2639 if (fetch_robust_entry(&pending, &head->list_op_pending, &pip))
2642 next_entry = NULL; /* avoid warning with gcc */
2643 while (entry != &head->list) {
2645 * Fetch the next entry in the list before calling
2646 * handle_futex_death:
2648 rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi);
2650 * A pending lock might already be on the list, so
2651 * don't process it twice:
2653 if (entry != pending)
2654 if (handle_futex_death((void __user *)entry + futex_offset,
2662 * Avoid excessively long or circular lists:
2671 handle_futex_death((void __user *)pending + futex_offset,
2675 long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout,
2676 u32 __user *uaddr2, u32 val2, u32 val3)
2678 int cmd = op & FUTEX_CMD_MASK;
2679 unsigned int flags = 0;
2681 if (!(op & FUTEX_PRIVATE_FLAG))
2682 flags |= FLAGS_SHARED;
2684 if (op & FUTEX_CLOCK_REALTIME) {
2685 flags |= FLAGS_CLOCKRT;
2686 if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI)
2692 case FUTEX_UNLOCK_PI:
2693 case FUTEX_TRYLOCK_PI:
2694 case FUTEX_WAIT_REQUEUE_PI:
2695 case FUTEX_CMP_REQUEUE_PI:
2696 if (!futex_cmpxchg_enabled)
2702 val3 = FUTEX_BITSET_MATCH_ANY;
2703 case FUTEX_WAIT_BITSET:
2704 return futex_wait(uaddr, flags, val, timeout, val3);
2706 val3 = FUTEX_BITSET_MATCH_ANY;
2707 case FUTEX_WAKE_BITSET:
2708 return futex_wake(uaddr, flags, val, val3);
2710 return futex_requeue(uaddr, flags, uaddr2, val, val2, NULL, 0);
2711 case FUTEX_CMP_REQUEUE:
2712 return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 0);
2714 return futex_wake_op(uaddr, flags, uaddr2, val, val2, val3);
2716 return futex_lock_pi(uaddr, flags, val, timeout, 0);
2717 case FUTEX_UNLOCK_PI:
2718 return futex_unlock_pi(uaddr, flags);
2719 case FUTEX_TRYLOCK_PI:
2720 return futex_lock_pi(uaddr, flags, 0, timeout, 1);
2721 case FUTEX_WAIT_REQUEUE_PI:
2722 val3 = FUTEX_BITSET_MATCH_ANY;
2723 return futex_wait_requeue_pi(uaddr, flags, val, timeout, val3,
2725 case FUTEX_CMP_REQUEUE_PI:
2726 return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 1);
2732 SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val,
2733 struct timespec __user *, utime, u32 __user *, uaddr2,
2737 ktime_t t, *tp = NULL;
2739 int cmd = op & FUTEX_CMD_MASK;
2741 if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI ||
2742 cmd == FUTEX_WAIT_BITSET ||
2743 cmd == FUTEX_WAIT_REQUEUE_PI)) {
2744 if (copy_from_user(&ts, utime, sizeof(ts)) != 0)
2746 if (!timespec_valid(&ts))
2749 t = timespec_to_ktime(ts);
2750 if (cmd == FUTEX_WAIT)
2751 t = ktime_add_safe(ktime_get(), t);
2755 * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*.
2756 * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP.
2758 if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE ||
2759 cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP)
2760 val2 = (u32) (unsigned long) utime;
2762 return do_futex(uaddr, op, val, tp, uaddr2, val2, val3);
2765 static int __init futex_init(void)
2771 * This will fail and we want it. Some arch implementations do
2772 * runtime detection of the futex_atomic_cmpxchg_inatomic()
2773 * functionality. We want to know that before we call in any
2774 * of the complex code paths. Also we want to prevent
2775 * registration of robust lists in that case. NULL is
2776 * guaranteed to fault and we get -EFAULT on functional
2777 * implementation, the non-functional ones will return
2780 if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT)
2781 futex_cmpxchg_enabled = 1;
2783 for (i = 0; i < ARRAY_SIZE(futex_queues); i++) {
2784 plist_head_init(&futex_queues[i].chain);
2785 spin_lock_init(&futex_queues[i].lock);
2790 __initcall(futex_init);