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/hugetlb.h>
65 #include <linux/freezer.h>
66 #include <linux/bootmem.h>
68 #include <asm/futex.h>
70 #include "locking/rtmutex_common.h"
73 * Basic futex operation and ordering guarantees:
75 * The waiter reads the futex value in user space and calls
76 * futex_wait(). This function computes the hash bucket and acquires
77 * the hash bucket lock. After that it reads the futex user space value
78 * again and verifies that the data has not changed. If it has not
79 * changed it enqueues itself into the hash bucket, releases the hash
80 * bucket lock and schedules.
82 * The waker side modifies the user space value of the futex and calls
83 * futex_wake(). This functions computes the hash bucket and acquires
84 * the hash bucket lock. Then it looks for waiters on that futex in the
85 * hash bucket and wakes them.
87 * Note that the spin_lock serializes waiters and wakers, so that the
88 * following scenario is avoided:
92 * sys_futex(WAIT, futex, val);
93 * futex_wait(futex, val);
96 * sys_futex(WAKE, futex);
101 * lock(hash_bucket(futex));
103 * unlock(hash_bucket(futex));
106 * This would cause the waiter on CPU 0 to wait forever because it
107 * missed the transition of the user space value from val to newval
108 * and the waker did not find the waiter in the hash bucket queue.
109 * The spinlock serializes that:
113 * sys_futex(WAIT, futex, val);
114 * futex_wait(futex, val);
115 * lock(hash_bucket(futex));
118 * sys_futex(WAKE, futex);
120 * lock(hash_bucket(futex));
123 * unlock(hash_bucket(futex));
124 * schedule(); if (!queue_empty())
125 * wake_waiters(futex);
126 * unlock(hash_bucket(futex));
129 int __read_mostly futex_cmpxchg_enabled;
132 * Futex flags used to encode options to functions and preserve them across
135 #define FLAGS_SHARED 0x01
136 #define FLAGS_CLOCKRT 0x02
137 #define FLAGS_HAS_TIMEOUT 0x04
140 * Priority Inheritance state:
142 struct futex_pi_state {
144 * list of 'owned' pi_state instances - these have to be
145 * cleaned up in do_exit() if the task exits prematurely:
147 struct list_head list;
152 struct rt_mutex pi_mutex;
154 struct task_struct *owner;
161 * struct futex_q - The hashed futex queue entry, one per waiting task
162 * @list: priority-sorted list of tasks waiting on this futex
163 * @task: the task waiting on the futex
164 * @lock_ptr: the hash bucket lock
165 * @key: the key the futex is hashed on
166 * @pi_state: optional priority inheritance state
167 * @rt_waiter: rt_waiter storage for use with requeue_pi
168 * @requeue_pi_key: the requeue_pi target futex key
169 * @bitset: bitset for the optional bitmasked wakeup
171 * We use this hashed waitqueue, instead of a normal wait_queue_t, so
172 * we can wake only the relevant ones (hashed queues may be shared).
174 * A futex_q has a woken state, just like tasks have TASK_RUNNING.
175 * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0.
176 * The order of wakeup is always to make the first condition true, then
179 * PI futexes are typically woken before they are removed from the hash list via
180 * the rt_mutex code. See unqueue_me_pi().
183 struct plist_node list;
185 struct task_struct *task;
186 spinlock_t *lock_ptr;
188 struct futex_pi_state *pi_state;
189 struct rt_mutex_waiter *rt_waiter;
190 union futex_key *requeue_pi_key;
194 static const struct futex_q futex_q_init = {
195 /* list gets initialized in queue_me()*/
196 .key = FUTEX_KEY_INIT,
197 .bitset = FUTEX_BITSET_MATCH_ANY
201 * Hash buckets are shared by all the futex_keys that hash to the same
202 * location. Each key may have multiple futex_q structures, one for each task
203 * waiting on a futex.
205 struct futex_hash_bucket {
207 struct plist_head chain;
208 } ____cacheline_aligned_in_smp;
210 static unsigned long __read_mostly futex_hashsize;
212 static struct futex_hash_bucket *futex_queues;
215 * We hash on the keys returned from get_futex_key (see below).
217 static struct futex_hash_bucket *hash_futex(union futex_key *key)
219 u32 hash = jhash2((u32*)&key->both.word,
220 (sizeof(key->both.word)+sizeof(key->both.ptr))/4,
222 return &futex_queues[hash & (futex_hashsize - 1)];
226 * Return 1 if two futex_keys are equal, 0 otherwise.
228 static inline int match_futex(union futex_key *key1, union futex_key *key2)
231 && key1->both.word == key2->both.word
232 && key1->both.ptr == key2->both.ptr
233 && key1->both.offset == key2->both.offset);
237 * Take a reference to the resource addressed by a key.
238 * Can be called while holding spinlocks.
241 static void get_futex_key_refs(union futex_key *key)
246 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
248 ihold(key->shared.inode);
250 case FUT_OFF_MMSHARED:
251 atomic_inc(&key->private.mm->mm_count);
257 * Drop a reference to the resource addressed by a key.
258 * The hash bucket spinlock must not be held.
260 static void drop_futex_key_refs(union futex_key *key)
262 if (!key->both.ptr) {
263 /* If we're here then we tried to put a key we failed to get */
268 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
270 iput(key->shared.inode);
272 case FUT_OFF_MMSHARED:
273 mmdrop(key->private.mm);
279 * get_futex_key() - Get parameters which are the keys for a futex
280 * @uaddr: virtual address of the futex
281 * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED
282 * @key: address where result is stored.
283 * @rw: mapping needs to be read/write (values: VERIFY_READ,
286 * Return: a negative error code or 0
288 * The key words are stored in *key on success.
290 * For shared mappings, it's (page->index, file_inode(vma->vm_file),
291 * offset_within_page). For private mappings, it's (uaddr, current->mm).
292 * We can usually work out the index without swapping in the page.
294 * lock_page() might sleep, the caller should not hold a spinlock.
297 get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key, int rw)
299 unsigned long address = (unsigned long)uaddr;
300 struct mm_struct *mm = current->mm;
301 struct page *page, *page_head;
305 * The futex address must be "naturally" aligned.
307 key->both.offset = address % PAGE_SIZE;
308 if (unlikely((address % sizeof(u32)) != 0))
310 address -= key->both.offset;
312 if (unlikely(!access_ok(rw, uaddr, sizeof(u32))))
316 * PROCESS_PRIVATE futexes are fast.
317 * As the mm cannot disappear under us and the 'key' only needs
318 * virtual address, we dont even have to find the underlying vma.
319 * Note : We do have to check 'uaddr' is a valid user address,
320 * but access_ok() should be faster than find_vma()
323 key->private.mm = mm;
324 key->private.address = address;
325 get_futex_key_refs(key);
330 err = get_user_pages_fast(address, 1, 1, &page);
332 * If write access is not required (eg. FUTEX_WAIT), try
333 * and get read-only access.
335 if (err == -EFAULT && rw == VERIFY_READ) {
336 err = get_user_pages_fast(address, 1, 0, &page);
344 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
346 if (unlikely(PageTail(page))) {
348 /* serialize against __split_huge_page_splitting() */
350 if (likely(__get_user_pages_fast(address, 1, !ro, &page) == 1)) {
351 page_head = compound_head(page);
353 * page_head is valid pointer but we must pin
354 * it before taking the PG_lock and/or
355 * PG_compound_lock. The moment we re-enable
356 * irqs __split_huge_page_splitting() can
357 * return and the head page can be freed from
358 * under us. We can't take the PG_lock and/or
359 * PG_compound_lock on a page that could be
360 * freed from under us.
362 if (page != page_head) {
373 page_head = compound_head(page);
374 if (page != page_head) {
380 lock_page(page_head);
383 * If page_head->mapping is NULL, then it cannot be a PageAnon
384 * page; but it might be the ZERO_PAGE or in the gate area or
385 * in a special mapping (all cases which we are happy to fail);
386 * or it may have been a good file page when get_user_pages_fast
387 * found it, but truncated or holepunched or subjected to
388 * invalidate_complete_page2 before we got the page lock (also
389 * cases which we are happy to fail). And we hold a reference,
390 * so refcount care in invalidate_complete_page's remove_mapping
391 * prevents drop_caches from setting mapping to NULL beneath us.
393 * The case we do have to guard against is when memory pressure made
394 * shmem_writepage move it from filecache to swapcache beneath us:
395 * an unlikely race, but we do need to retry for page_head->mapping.
397 if (!page_head->mapping) {
398 int shmem_swizzled = PageSwapCache(page_head);
399 unlock_page(page_head);
407 * Private mappings are handled in a simple way.
409 * NOTE: When userspace waits on a MAP_SHARED mapping, even if
410 * it's a read-only handle, it's expected that futexes attach to
411 * the object not the particular process.
413 if (PageAnon(page_head)) {
415 * A RO anonymous page will never change and thus doesn't make
416 * sense for futex operations.
423 key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */
424 key->private.mm = mm;
425 key->private.address = address;
427 key->both.offset |= FUT_OFF_INODE; /* inode-based key */
428 key->shared.inode = page_head->mapping->host;
429 key->shared.pgoff = basepage_index(page);
432 get_futex_key_refs(key);
435 unlock_page(page_head);
440 static inline void put_futex_key(union futex_key *key)
442 drop_futex_key_refs(key);
446 * fault_in_user_writeable() - Fault in user address and verify RW access
447 * @uaddr: pointer to faulting user space address
449 * Slow path to fixup the fault we just took in the atomic write
452 * We have no generic implementation of a non-destructive write to the
453 * user address. We know that we faulted in the atomic pagefault
454 * disabled section so we can as well avoid the #PF overhead by
455 * calling get_user_pages() right away.
457 static int fault_in_user_writeable(u32 __user *uaddr)
459 struct mm_struct *mm = current->mm;
462 down_read(&mm->mmap_sem);
463 ret = fixup_user_fault(current, mm, (unsigned long)uaddr,
465 up_read(&mm->mmap_sem);
467 return ret < 0 ? ret : 0;
471 * futex_top_waiter() - Return the highest priority waiter on a futex
472 * @hb: the hash bucket the futex_q's reside in
473 * @key: the futex key (to distinguish it from other futex futex_q's)
475 * Must be called with the hb lock held.
477 static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb,
478 union futex_key *key)
480 struct futex_q *this;
482 plist_for_each_entry(this, &hb->chain, list) {
483 if (match_futex(&this->key, key))
489 static int cmpxchg_futex_value_locked(u32 *curval, u32 __user *uaddr,
490 u32 uval, u32 newval)
495 ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval);
501 static int get_futex_value_locked(u32 *dest, u32 __user *from)
506 ret = __copy_from_user_inatomic(dest, from, sizeof(u32));
509 return ret ? -EFAULT : 0;
516 static int refill_pi_state_cache(void)
518 struct futex_pi_state *pi_state;
520 if (likely(current->pi_state_cache))
523 pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL);
528 INIT_LIST_HEAD(&pi_state->list);
529 /* pi_mutex gets initialized later */
530 pi_state->owner = NULL;
531 atomic_set(&pi_state->refcount, 1);
532 pi_state->key = FUTEX_KEY_INIT;
534 current->pi_state_cache = pi_state;
539 static struct futex_pi_state * alloc_pi_state(void)
541 struct futex_pi_state *pi_state = current->pi_state_cache;
544 current->pi_state_cache = NULL;
549 static void free_pi_state(struct futex_pi_state *pi_state)
551 if (!atomic_dec_and_test(&pi_state->refcount))
555 * If pi_state->owner is NULL, the owner is most probably dying
556 * and has cleaned up the pi_state already
558 if (pi_state->owner) {
559 raw_spin_lock_irq(&pi_state->owner->pi_lock);
560 list_del_init(&pi_state->list);
561 raw_spin_unlock_irq(&pi_state->owner->pi_lock);
563 rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner);
566 if (current->pi_state_cache)
570 * pi_state->list is already empty.
571 * clear pi_state->owner.
572 * refcount is at 0 - put it back to 1.
574 pi_state->owner = NULL;
575 atomic_set(&pi_state->refcount, 1);
576 current->pi_state_cache = pi_state;
581 * Look up the task based on what TID userspace gave us.
584 static struct task_struct * futex_find_get_task(pid_t pid)
586 struct task_struct *p;
589 p = find_task_by_vpid(pid);
599 * This task is holding PI mutexes at exit time => bad.
600 * Kernel cleans up PI-state, but userspace is likely hosed.
601 * (Robust-futex cleanup is separate and might save the day for userspace.)
603 void exit_pi_state_list(struct task_struct *curr)
605 struct list_head *next, *head = &curr->pi_state_list;
606 struct futex_pi_state *pi_state;
607 struct futex_hash_bucket *hb;
608 union futex_key key = FUTEX_KEY_INIT;
610 if (!futex_cmpxchg_enabled)
613 * We are a ZOMBIE and nobody can enqueue itself on
614 * pi_state_list anymore, but we have to be careful
615 * versus waiters unqueueing themselves:
617 raw_spin_lock_irq(&curr->pi_lock);
618 while (!list_empty(head)) {
621 pi_state = list_entry(next, struct futex_pi_state, list);
623 hb = hash_futex(&key);
624 raw_spin_unlock_irq(&curr->pi_lock);
626 spin_lock(&hb->lock);
628 raw_spin_lock_irq(&curr->pi_lock);
630 * We dropped the pi-lock, so re-check whether this
631 * task still owns the PI-state:
633 if (head->next != next) {
634 spin_unlock(&hb->lock);
638 WARN_ON(pi_state->owner != curr);
639 WARN_ON(list_empty(&pi_state->list));
640 list_del_init(&pi_state->list);
641 pi_state->owner = NULL;
642 raw_spin_unlock_irq(&curr->pi_lock);
644 rt_mutex_unlock(&pi_state->pi_mutex);
646 spin_unlock(&hb->lock);
648 raw_spin_lock_irq(&curr->pi_lock);
650 raw_spin_unlock_irq(&curr->pi_lock);
654 lookup_pi_state(u32 uval, struct futex_hash_bucket *hb,
655 union futex_key *key, struct futex_pi_state **ps)
657 struct futex_pi_state *pi_state = NULL;
658 struct futex_q *this, *next;
659 struct task_struct *p;
660 pid_t pid = uval & FUTEX_TID_MASK;
662 plist_for_each_entry_safe(this, next, &hb->chain, list) {
663 if (match_futex(&this->key, key)) {
665 * Another waiter already exists - bump up
666 * the refcount and return its pi_state:
668 pi_state = this->pi_state;
670 * Userspace might have messed up non-PI and PI futexes
672 if (unlikely(!pi_state))
675 WARN_ON(!atomic_read(&pi_state->refcount));
678 * When pi_state->owner is NULL then the owner died
679 * and another waiter is on the fly. pi_state->owner
680 * is fixed up by the task which acquires
681 * pi_state->rt_mutex.
683 * We do not check for pid == 0 which can happen when
684 * the owner died and robust_list_exit() cleared the
687 if (pid && pi_state->owner) {
689 * Bail out if user space manipulated the
692 if (pid != task_pid_vnr(pi_state->owner))
696 atomic_inc(&pi_state->refcount);
704 * We are the first waiter - try to look up the real owner and attach
705 * the new pi_state to it, but bail out when TID = 0
709 p = futex_find_get_task(pid);
714 * We need to look at the task state flags to figure out,
715 * whether the task is exiting. To protect against the do_exit
716 * change of the task flags, we do this protected by
719 raw_spin_lock_irq(&p->pi_lock);
720 if (unlikely(p->flags & PF_EXITING)) {
722 * The task is on the way out. When PF_EXITPIDONE is
723 * set, we know that the task has finished the
726 int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN;
728 raw_spin_unlock_irq(&p->pi_lock);
733 pi_state = alloc_pi_state();
736 * Initialize the pi_mutex in locked state and make 'p'
739 rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p);
741 /* Store the key for possible exit cleanups: */
742 pi_state->key = *key;
744 WARN_ON(!list_empty(&pi_state->list));
745 list_add(&pi_state->list, &p->pi_state_list);
747 raw_spin_unlock_irq(&p->pi_lock);
757 * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex
758 * @uaddr: the pi futex user address
759 * @hb: the pi futex hash bucket
760 * @key: the futex key associated with uaddr and hb
761 * @ps: the pi_state pointer where we store the result of the
763 * @task: the task to perform the atomic lock work for. This will
764 * be "current" except in the case of requeue pi.
765 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
769 * 1 - acquired the lock;
772 * The hb->lock and futex_key refs shall be held by the caller.
774 static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb,
775 union futex_key *key,
776 struct futex_pi_state **ps,
777 struct task_struct *task, int set_waiters)
779 int lock_taken, ret, force_take = 0;
780 u32 uval, newval, curval, vpid = task_pid_vnr(task);
783 ret = lock_taken = 0;
786 * To avoid races, we attempt to take the lock here again
787 * (by doing a 0 -> TID atomic cmpxchg), while holding all
788 * the locks. It will most likely not succeed.
792 newval |= FUTEX_WAITERS;
794 if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, 0, newval)))
800 if ((unlikely((curval & FUTEX_TID_MASK) == vpid)))
804 * Surprise - we got the lock. Just return to userspace:
806 if (unlikely(!curval))
812 * Set the FUTEX_WAITERS flag, so the owner will know it has someone
813 * to wake at the next unlock.
815 newval = curval | FUTEX_WAITERS;
818 * Should we force take the futex? See below.
820 if (unlikely(force_take)) {
822 * Keep the OWNER_DIED and the WAITERS bit and set the
825 newval = (curval & ~FUTEX_TID_MASK) | vpid;
830 if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)))
832 if (unlikely(curval != uval))
836 * We took the lock due to forced take over.
838 if (unlikely(lock_taken))
842 * We dont have the lock. Look up the PI state (or create it if
843 * we are the first waiter):
845 ret = lookup_pi_state(uval, hb, key, ps);
851 * We failed to find an owner for this
852 * futex. So we have no pi_state to block
853 * on. This can happen in two cases:
856 * 2) A stale FUTEX_WAITERS bit
858 * Re-read the futex value.
860 if (get_futex_value_locked(&curval, uaddr))
864 * If the owner died or we have a stale
865 * WAITERS bit the owner TID in the user space
868 if (!(curval & FUTEX_TID_MASK)) {
881 * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket
882 * @q: The futex_q to unqueue
884 * The q->lock_ptr must not be NULL and must be held by the caller.
886 static void __unqueue_futex(struct futex_q *q)
888 struct futex_hash_bucket *hb;
890 if (WARN_ON_SMP(!q->lock_ptr || !spin_is_locked(q->lock_ptr))
891 || WARN_ON(plist_node_empty(&q->list)))
894 hb = container_of(q->lock_ptr, struct futex_hash_bucket, lock);
895 plist_del(&q->list, &hb->chain);
899 * The hash bucket lock must be held when this is called.
900 * Afterwards, the futex_q must not be accessed.
902 static void wake_futex(struct futex_q *q)
904 struct task_struct *p = q->task;
906 if (WARN(q->pi_state || q->rt_waiter, "refusing to wake PI futex\n"))
910 * We set q->lock_ptr = NULL _before_ we wake up the task. If
911 * a non-futex wake up happens on another CPU then the task
912 * might exit and p would dereference a non-existing task
913 * struct. Prevent this by holding a reference on p across the
920 * The waiting task can free the futex_q as soon as
921 * q->lock_ptr = NULL is written, without taking any locks. A
922 * memory barrier is required here to prevent the following
923 * store to lock_ptr from getting ahead of the plist_del.
928 wake_up_state(p, TASK_NORMAL);
932 static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this)
934 struct task_struct *new_owner;
935 struct futex_pi_state *pi_state = this->pi_state;
936 u32 uninitialized_var(curval), newval;
942 * If current does not own the pi_state then the futex is
943 * inconsistent and user space fiddled with the futex value.
945 if (pi_state->owner != current)
948 raw_spin_lock(&pi_state->pi_mutex.wait_lock);
949 new_owner = rt_mutex_next_owner(&pi_state->pi_mutex);
952 * It is possible that the next waiter (the one that brought
953 * this owner to the kernel) timed out and is no longer
954 * waiting on the lock.
957 new_owner = this->task;
960 * We pass it to the next owner. (The WAITERS bit is always
961 * kept enabled while there is PI state around. We must also
962 * preserve the owner died bit.)
964 if (!(uval & FUTEX_OWNER_DIED)) {
967 newval = FUTEX_WAITERS | task_pid_vnr(new_owner);
969 if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))
971 else if (curval != uval)
974 raw_spin_unlock(&pi_state->pi_mutex.wait_lock);
979 raw_spin_lock_irq(&pi_state->owner->pi_lock);
980 WARN_ON(list_empty(&pi_state->list));
981 list_del_init(&pi_state->list);
982 raw_spin_unlock_irq(&pi_state->owner->pi_lock);
984 raw_spin_lock_irq(&new_owner->pi_lock);
985 WARN_ON(!list_empty(&pi_state->list));
986 list_add(&pi_state->list, &new_owner->pi_state_list);
987 pi_state->owner = new_owner;
988 raw_spin_unlock_irq(&new_owner->pi_lock);
990 raw_spin_unlock(&pi_state->pi_mutex.wait_lock);
991 rt_mutex_unlock(&pi_state->pi_mutex);
996 static int unlock_futex_pi(u32 __user *uaddr, u32 uval)
998 u32 uninitialized_var(oldval);
1001 * There is no waiter, so we unlock the futex. The owner died
1002 * bit has not to be preserved here. We are the owner:
1004 if (cmpxchg_futex_value_locked(&oldval, uaddr, uval, 0))
1013 * Express the locking dependencies for lockdep:
1016 double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
1019 spin_lock(&hb1->lock);
1021 spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING);
1022 } else { /* hb1 > hb2 */
1023 spin_lock(&hb2->lock);
1024 spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING);
1029 double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
1031 spin_unlock(&hb1->lock);
1033 spin_unlock(&hb2->lock);
1037 * Wake up waiters matching bitset queued on this futex (uaddr).
1040 futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset)
1042 struct futex_hash_bucket *hb;
1043 struct futex_q *this, *next;
1044 union futex_key key = FUTEX_KEY_INIT;
1050 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_READ);
1051 if (unlikely(ret != 0))
1054 hb = hash_futex(&key);
1055 spin_lock(&hb->lock);
1057 plist_for_each_entry_safe(this, next, &hb->chain, list) {
1058 if (match_futex (&this->key, &key)) {
1059 if (this->pi_state || this->rt_waiter) {
1064 /* Check if one of the bits is set in both bitsets */
1065 if (!(this->bitset & bitset))
1069 if (++ret >= nr_wake)
1074 spin_unlock(&hb->lock);
1075 put_futex_key(&key);
1081 * Wake up all waiters hashed on the physical page that is mapped
1082 * to this virtual address:
1085 futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2,
1086 int nr_wake, int nr_wake2, int op)
1088 union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
1089 struct futex_hash_bucket *hb1, *hb2;
1090 struct futex_q *this, *next;
1094 ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ);
1095 if (unlikely(ret != 0))
1097 ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE);
1098 if (unlikely(ret != 0))
1101 hb1 = hash_futex(&key1);
1102 hb2 = hash_futex(&key2);
1105 double_lock_hb(hb1, hb2);
1106 op_ret = futex_atomic_op_inuser(op, uaddr2);
1107 if (unlikely(op_ret < 0)) {
1109 double_unlock_hb(hb1, hb2);
1113 * we don't get EFAULT from MMU faults if we don't have an MMU,
1114 * but we might get them from range checking
1120 if (unlikely(op_ret != -EFAULT)) {
1125 ret = fault_in_user_writeable(uaddr2);
1129 if (!(flags & FLAGS_SHARED))
1132 put_futex_key(&key2);
1133 put_futex_key(&key1);
1137 plist_for_each_entry_safe(this, next, &hb1->chain, list) {
1138 if (match_futex (&this->key, &key1)) {
1139 if (this->pi_state || this->rt_waiter) {
1144 if (++ret >= nr_wake)
1151 plist_for_each_entry_safe(this, next, &hb2->chain, list) {
1152 if (match_futex (&this->key, &key2)) {
1153 if (this->pi_state || this->rt_waiter) {
1158 if (++op_ret >= nr_wake2)
1166 double_unlock_hb(hb1, hb2);
1168 put_futex_key(&key2);
1170 put_futex_key(&key1);
1176 * requeue_futex() - Requeue a futex_q from one hb to another
1177 * @q: the futex_q to requeue
1178 * @hb1: the source hash_bucket
1179 * @hb2: the target hash_bucket
1180 * @key2: the new key for the requeued futex_q
1183 void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1,
1184 struct futex_hash_bucket *hb2, union futex_key *key2)
1188 * If key1 and key2 hash to the same bucket, no need to
1191 if (likely(&hb1->chain != &hb2->chain)) {
1192 plist_del(&q->list, &hb1->chain);
1193 plist_add(&q->list, &hb2->chain);
1194 q->lock_ptr = &hb2->lock;
1196 get_futex_key_refs(key2);
1201 * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue
1203 * @key: the key of the requeue target futex
1204 * @hb: the hash_bucket of the requeue target futex
1206 * During futex_requeue, with requeue_pi=1, it is possible to acquire the
1207 * target futex if it is uncontended or via a lock steal. Set the futex_q key
1208 * to the requeue target futex so the waiter can detect the wakeup on the right
1209 * futex, but remove it from the hb and NULL the rt_waiter so it can detect
1210 * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock
1211 * to protect access to the pi_state to fixup the owner later. Must be called
1212 * with both q->lock_ptr and hb->lock held.
1215 void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key,
1216 struct futex_hash_bucket *hb)
1218 get_futex_key_refs(key);
1223 WARN_ON(!q->rt_waiter);
1224 q->rt_waiter = NULL;
1226 q->lock_ptr = &hb->lock;
1228 wake_up_state(q->task, TASK_NORMAL);
1232 * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter
1233 * @pifutex: the user address of the to futex
1234 * @hb1: the from futex hash bucket, must be locked by the caller
1235 * @hb2: the to futex hash bucket, must be locked by the caller
1236 * @key1: the from futex key
1237 * @key2: the to futex key
1238 * @ps: address to store the pi_state pointer
1239 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
1241 * Try and get the lock on behalf of the top waiter if we can do it atomically.
1242 * Wake the top waiter if we succeed. If the caller specified set_waiters,
1243 * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit.
1244 * hb1 and hb2 must be held by the caller.
1247 * 0 - failed to acquire the lock atomically;
1248 * 1 - acquired the lock;
1251 static int futex_proxy_trylock_atomic(u32 __user *pifutex,
1252 struct futex_hash_bucket *hb1,
1253 struct futex_hash_bucket *hb2,
1254 union futex_key *key1, union futex_key *key2,
1255 struct futex_pi_state **ps, int set_waiters)
1257 struct futex_q *top_waiter = NULL;
1261 if (get_futex_value_locked(&curval, pifutex))
1265 * Find the top_waiter and determine if there are additional waiters.
1266 * If the caller intends to requeue more than 1 waiter to pifutex,
1267 * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now,
1268 * as we have means to handle the possible fault. If not, don't set
1269 * the bit unecessarily as it will force the subsequent unlock to enter
1272 top_waiter = futex_top_waiter(hb1, key1);
1274 /* There are no waiters, nothing for us to do. */
1278 /* Ensure we requeue to the expected futex. */
1279 if (!match_futex(top_waiter->requeue_pi_key, key2))
1283 * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in
1284 * the contended case or if set_waiters is 1. The pi_state is returned
1285 * in ps in contended cases.
1287 ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task,
1290 requeue_pi_wake_futex(top_waiter, key2, hb2);
1296 * futex_requeue() - Requeue waiters from uaddr1 to uaddr2
1297 * @uaddr1: source futex user address
1298 * @flags: futex flags (FLAGS_SHARED, etc.)
1299 * @uaddr2: target futex user address
1300 * @nr_wake: number of waiters to wake (must be 1 for requeue_pi)
1301 * @nr_requeue: number of waiters to requeue (0-INT_MAX)
1302 * @cmpval: @uaddr1 expected value (or %NULL)
1303 * @requeue_pi: if we are attempting to requeue from a non-pi futex to a
1304 * pi futex (pi to pi requeue is not supported)
1306 * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire
1307 * uaddr2 atomically on behalf of the top waiter.
1310 * >=0 - on success, the number of tasks requeued or woken;
1313 static int futex_requeue(u32 __user *uaddr1, unsigned int flags,
1314 u32 __user *uaddr2, int nr_wake, int nr_requeue,
1315 u32 *cmpval, int requeue_pi)
1317 union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
1318 int drop_count = 0, task_count = 0, ret;
1319 struct futex_pi_state *pi_state = NULL;
1320 struct futex_hash_bucket *hb1, *hb2;
1321 struct futex_q *this, *next;
1326 * requeue_pi requires a pi_state, try to allocate it now
1327 * without any locks in case it fails.
1329 if (refill_pi_state_cache())
1332 * requeue_pi must wake as many tasks as it can, up to nr_wake
1333 * + nr_requeue, since it acquires the rt_mutex prior to
1334 * returning to userspace, so as to not leave the rt_mutex with
1335 * waiters and no owner. However, second and third wake-ups
1336 * cannot be predicted as they involve race conditions with the
1337 * first wake and a fault while looking up the pi_state. Both
1338 * pthread_cond_signal() and pthread_cond_broadcast() should
1346 if (pi_state != NULL) {
1348 * We will have to lookup the pi_state again, so free this one
1349 * to keep the accounting correct.
1351 free_pi_state(pi_state);
1355 ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ);
1356 if (unlikely(ret != 0))
1358 ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2,
1359 requeue_pi ? VERIFY_WRITE : VERIFY_READ);
1360 if (unlikely(ret != 0))
1363 hb1 = hash_futex(&key1);
1364 hb2 = hash_futex(&key2);
1367 double_lock_hb(hb1, hb2);
1369 if (likely(cmpval != NULL)) {
1372 ret = get_futex_value_locked(&curval, uaddr1);
1374 if (unlikely(ret)) {
1375 double_unlock_hb(hb1, hb2);
1377 ret = get_user(curval, uaddr1);
1381 if (!(flags & FLAGS_SHARED))
1384 put_futex_key(&key2);
1385 put_futex_key(&key1);
1388 if (curval != *cmpval) {
1394 if (requeue_pi && (task_count - nr_wake < nr_requeue)) {
1396 * Attempt to acquire uaddr2 and wake the top waiter. If we
1397 * intend to requeue waiters, force setting the FUTEX_WAITERS
1398 * bit. We force this here where we are able to easily handle
1399 * faults rather in the requeue loop below.
1401 ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1,
1402 &key2, &pi_state, nr_requeue);
1405 * At this point the top_waiter has either taken uaddr2 or is
1406 * waiting on it. If the former, then the pi_state will not
1407 * exist yet, look it up one more time to ensure we have a
1414 ret = get_futex_value_locked(&curval2, uaddr2);
1416 ret = lookup_pi_state(curval2, hb2, &key2,
1424 double_unlock_hb(hb1, hb2);
1425 put_futex_key(&key2);
1426 put_futex_key(&key1);
1427 ret = fault_in_user_writeable(uaddr2);
1432 /* The owner was exiting, try again. */
1433 double_unlock_hb(hb1, hb2);
1434 put_futex_key(&key2);
1435 put_futex_key(&key1);
1443 plist_for_each_entry_safe(this, next, &hb1->chain, 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_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)
1859 freezable_schedule();
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);
1918 ret = get_user(uval, uaddr);
1922 if (!(flags & FLAGS_SHARED))
1925 put_futex_key(&q->key);
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
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 */
2133 put_futex_key(&q.key);
2136 destroy_hrtimer_on_stack(&to->timer);
2137 return ret != -EINTR ? ret : -ERESTARTNOINTR;
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 union futex_key key = FUTEX_KEY_INIT;
2163 u32 uval, vpid = task_pid_vnr(current);
2167 if (get_user(uval, uaddr))
2170 * We release only a lock we actually own:
2172 if ((uval & FUTEX_TID_MASK) != vpid)
2175 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_WRITE);
2176 if (unlikely(ret != 0))
2179 hb = hash_futex(&key);
2180 spin_lock(&hb->lock);
2183 * To avoid races, try to do the TID -> 0 atomic transition
2184 * again. If it succeeds then we can return without waking
2187 if (!(uval & FUTEX_OWNER_DIED) &&
2188 cmpxchg_futex_value_locked(&uval, uaddr, vpid, 0))
2191 * Rare case: we managed to release the lock atomically,
2192 * no need to wake anyone else up:
2194 if (unlikely(uval == vpid))
2198 * Ok, other tasks may need to be woken up - check waiters
2199 * and do the wakeup if necessary:
2201 plist_for_each_entry_safe(this, next, &hb->chain, list) {
2202 if (!match_futex (&this->key, &key))
2204 ret = wake_futex_pi(uaddr, uval, this);
2206 * The atomic access to the futex value
2207 * generated a pagefault, so retry the
2208 * user-access and the wakeup:
2215 * No waiters - kernel unlocks the futex:
2217 if (!(uval & FUTEX_OWNER_DIED)) {
2218 ret = unlock_futex_pi(uaddr, uval);
2224 spin_unlock(&hb->lock);
2225 put_futex_key(&key);
2231 spin_unlock(&hb->lock);
2232 put_futex_key(&key);
2234 ret = fault_in_user_writeable(uaddr);
2242 * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex
2243 * @hb: the hash_bucket futex_q was original enqueued on
2244 * @q: the futex_q woken while waiting to be requeued
2245 * @key2: the futex_key of the requeue target futex
2246 * @timeout: the timeout associated with the wait (NULL if none)
2248 * Detect if the task was woken on the initial futex as opposed to the requeue
2249 * target futex. If so, determine if it was a timeout or a signal that caused
2250 * the wakeup and return the appropriate error code to the caller. Must be
2251 * called with the hb lock held.
2254 * 0 = no early wakeup detected;
2255 * <0 = -ETIMEDOUT or -ERESTARTNOINTR
2258 int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb,
2259 struct futex_q *q, union futex_key *key2,
2260 struct hrtimer_sleeper *timeout)
2265 * With the hb lock held, we avoid races while we process the wakeup.
2266 * We only need to hold hb (and not hb2) to ensure atomicity as the
2267 * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb.
2268 * It can't be requeued from uaddr2 to something else since we don't
2269 * support a PI aware source futex for requeue.
2271 if (!match_futex(&q->key, key2)) {
2272 WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr));
2274 * We were woken prior to requeue by a timeout or a signal.
2275 * Unqueue the futex_q and determine which it was.
2277 plist_del(&q->list, &hb->chain);
2279 /* Handle spurious wakeups gracefully */
2281 if (timeout && !timeout->task)
2283 else if (signal_pending(current))
2284 ret = -ERESTARTNOINTR;
2290 * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2
2291 * @uaddr: the futex we initially wait on (non-pi)
2292 * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be
2293 * the same type, no requeueing from private to shared, etc.
2294 * @val: the expected value of uaddr
2295 * @abs_time: absolute timeout
2296 * @bitset: 32 bit wakeup bitset set by userspace, defaults to all
2297 * @uaddr2: the pi futex we will take prior to returning to user-space
2299 * The caller will wait on uaddr and will be requeued by futex_requeue() to
2300 * uaddr2 which must be PI aware and unique from uaddr. Normal wakeup will wake
2301 * on uaddr2 and complete the acquisition of the rt_mutex prior to returning to
2302 * userspace. This ensures the rt_mutex maintains an owner when it has waiters;
2303 * without one, the pi logic would not know which task to boost/deboost, if
2304 * there was a need to.
2306 * We call schedule in futex_wait_queue_me() when we enqueue and return there
2307 * via the following--
2308 * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue()
2309 * 2) wakeup on uaddr2 after a requeue
2313 * If 3, cleanup and return -ERESTARTNOINTR.
2315 * If 2, we may then block on trying to take the rt_mutex and return via:
2316 * 5) successful lock
2319 * 8) other lock acquisition failure
2321 * If 6, return -EWOULDBLOCK (restarting the syscall would do the same).
2323 * If 4 or 7, we cleanup and return with -ETIMEDOUT.
2329 static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
2330 u32 val, ktime_t *abs_time, u32 bitset,
2333 struct hrtimer_sleeper timeout, *to = NULL;
2334 struct rt_mutex_waiter rt_waiter;
2335 struct rt_mutex *pi_mutex = NULL;
2336 struct futex_hash_bucket *hb;
2337 union futex_key key2 = FUTEX_KEY_INIT;
2338 struct futex_q q = futex_q_init;
2341 if (uaddr == uaddr2)
2349 hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ?
2350 CLOCK_REALTIME : CLOCK_MONOTONIC,
2352 hrtimer_init_sleeper(to, current);
2353 hrtimer_set_expires_range_ns(&to->timer, *abs_time,
2354 current->timer_slack_ns);
2358 * The waiter is allocated on our stack, manipulated by the requeue
2359 * code while we sleep on uaddr.
2361 debug_rt_mutex_init_waiter(&rt_waiter);
2362 rt_waiter.task = NULL;
2364 ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE);
2365 if (unlikely(ret != 0))
2369 q.rt_waiter = &rt_waiter;
2370 q.requeue_pi_key = &key2;
2373 * Prepare to wait on uaddr. On success, increments q.key (key1) ref
2376 ret = futex_wait_setup(uaddr, val, flags, &q, &hb);
2380 /* Queue the futex_q, drop the hb lock, wait for wakeup. */
2381 futex_wait_queue_me(hb, &q, to);
2383 spin_lock(&hb->lock);
2384 ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to);
2385 spin_unlock(&hb->lock);
2390 * In order for us to be here, we know our q.key == key2, and since
2391 * we took the hb->lock above, we also know that futex_requeue() has
2392 * completed and we no longer have to concern ourselves with a wakeup
2393 * race with the atomic proxy lock acquisition by the requeue code. The
2394 * futex_requeue dropped our key1 reference and incremented our key2
2398 /* Check if the requeue code acquired the second futex for us. */
2401 * Got the lock. We might not be the anticipated owner if we
2402 * did a lock-steal - fix up the PI-state in that case.
2404 if (q.pi_state && (q.pi_state->owner != current)) {
2405 spin_lock(q.lock_ptr);
2406 ret = fixup_pi_state_owner(uaddr2, &q, current);
2407 spin_unlock(q.lock_ptr);
2411 * We have been woken up by futex_unlock_pi(), a timeout, or a
2412 * signal. futex_unlock_pi() will not destroy the lock_ptr nor
2415 WARN_ON(!q.pi_state);
2416 pi_mutex = &q.pi_state->pi_mutex;
2417 ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter, 1);
2418 debug_rt_mutex_free_waiter(&rt_waiter);
2420 spin_lock(q.lock_ptr);
2422 * Fixup the pi_state owner and possibly acquire the lock if we
2425 res = fixup_owner(uaddr2, &q, !ret);
2427 * If fixup_owner() returned an error, proprogate that. If it
2428 * acquired the lock, clear -ETIMEDOUT or -EINTR.
2431 ret = (res < 0) ? res : 0;
2433 /* Unqueue and drop the lock. */
2438 * If fixup_pi_state_owner() faulted and was unable to handle the
2439 * fault, unlock the rt_mutex and return the fault to userspace.
2441 if (ret == -EFAULT) {
2442 if (pi_mutex && rt_mutex_owner(pi_mutex) == current)
2443 rt_mutex_unlock(pi_mutex);
2444 } else if (ret == -EINTR) {
2446 * We've already been requeued, but cannot restart by calling
2447 * futex_lock_pi() directly. We could restart this syscall, but
2448 * it would detect that the user space "val" changed and return
2449 * -EWOULDBLOCK. Save the overhead of the restart and return
2450 * -EWOULDBLOCK directly.
2456 put_futex_key(&q.key);
2458 put_futex_key(&key2);
2462 hrtimer_cancel(&to->timer);
2463 destroy_hrtimer_on_stack(&to->timer);
2469 * Support for robust futexes: the kernel cleans up held futexes at
2472 * Implementation: user-space maintains a per-thread list of locks it
2473 * is holding. Upon do_exit(), the kernel carefully walks this list,
2474 * and marks all locks that are owned by this thread with the
2475 * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is
2476 * always manipulated with the lock held, so the list is private and
2477 * per-thread. Userspace also maintains a per-thread 'list_op_pending'
2478 * field, to allow the kernel to clean up if the thread dies after
2479 * acquiring the lock, but just before it could have added itself to
2480 * the list. There can only be one such pending lock.
2484 * sys_set_robust_list() - Set the robust-futex list head of a task
2485 * @head: pointer to the list-head
2486 * @len: length of the list-head, as userspace expects
2488 SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head,
2491 if (!futex_cmpxchg_enabled)
2494 * The kernel knows only one size for now:
2496 if (unlikely(len != sizeof(*head)))
2499 current->robust_list = head;
2505 * sys_get_robust_list() - Get the robust-futex list head of a task
2506 * @pid: pid of the process [zero for current task]
2507 * @head_ptr: pointer to a list-head pointer, the kernel fills it in
2508 * @len_ptr: pointer to a length field, the kernel fills in the header size
2510 SYSCALL_DEFINE3(get_robust_list, int, pid,
2511 struct robust_list_head __user * __user *, head_ptr,
2512 size_t __user *, len_ptr)
2514 struct robust_list_head __user *head;
2516 struct task_struct *p;
2518 if (!futex_cmpxchg_enabled)
2527 p = find_task_by_vpid(pid);
2533 if (!ptrace_may_access(p, PTRACE_MODE_READ))
2536 head = p->robust_list;
2539 if (put_user(sizeof(*head), len_ptr))
2541 return put_user(head, head_ptr);
2550 * Process a futex-list entry, check whether it's owned by the
2551 * dying task, and do notification if so:
2553 int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi)
2555 u32 uval, uninitialized_var(nval), mval;
2558 if (get_user(uval, uaddr))
2561 if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) {
2563 * Ok, this dying thread is truly holding a futex
2564 * of interest. Set the OWNER_DIED bit atomically
2565 * via cmpxchg, and if the value had FUTEX_WAITERS
2566 * set, wake up a waiter (if any). (We have to do a
2567 * futex_wake() even if OWNER_DIED is already set -
2568 * to handle the rare but possible case of recursive
2569 * thread-death.) The rest of the cleanup is done in
2572 mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED;
2574 * We are not holding a lock here, but we want to have
2575 * the pagefault_disable/enable() protection because
2576 * we want to handle the fault gracefully. If the
2577 * access fails we try to fault in the futex with R/W
2578 * verification via get_user_pages. get_user() above
2579 * does not guarantee R/W access. If that fails we
2580 * give up and leave the futex locked.
2582 if (cmpxchg_futex_value_locked(&nval, uaddr, uval, mval)) {
2583 if (fault_in_user_writeable(uaddr))
2591 * Wake robust non-PI futexes here. The wakeup of
2592 * PI futexes happens in exit_pi_state():
2594 if (!pi && (uval & FUTEX_WAITERS))
2595 futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY);
2601 * Fetch a robust-list pointer. Bit 0 signals PI futexes:
2603 static inline int fetch_robust_entry(struct robust_list __user **entry,
2604 struct robust_list __user * __user *head,
2607 unsigned long uentry;
2609 if (get_user(uentry, (unsigned long __user *)head))
2612 *entry = (void __user *)(uentry & ~1UL);
2619 * Walk curr->robust_list (very carefully, it's a userspace list!)
2620 * and mark any locks found there dead, and notify any waiters.
2622 * We silently return on any sign of list-walking problem.
2624 void exit_robust_list(struct task_struct *curr)
2626 struct robust_list_head __user *head = curr->robust_list;
2627 struct robust_list __user *entry, *next_entry, *pending;
2628 unsigned int limit = ROBUST_LIST_LIMIT, pi, pip;
2629 unsigned int uninitialized_var(next_pi);
2630 unsigned long futex_offset;
2633 if (!futex_cmpxchg_enabled)
2637 * Fetch the list head (which was registered earlier, via
2638 * sys_set_robust_list()):
2640 if (fetch_robust_entry(&entry, &head->list.next, &pi))
2643 * Fetch the relative futex offset:
2645 if (get_user(futex_offset, &head->futex_offset))
2648 * Fetch any possibly pending lock-add first, and handle it
2651 if (fetch_robust_entry(&pending, &head->list_op_pending, &pip))
2654 next_entry = NULL; /* avoid warning with gcc */
2655 while (entry != &head->list) {
2657 * Fetch the next entry in the list before calling
2658 * handle_futex_death:
2660 rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi);
2662 * A pending lock might already be on the list, so
2663 * don't process it twice:
2665 if (entry != pending)
2666 if (handle_futex_death((void __user *)entry + futex_offset,
2674 * Avoid excessively long or circular lists:
2683 handle_futex_death((void __user *)pending + futex_offset,
2687 long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout,
2688 u32 __user *uaddr2, u32 val2, u32 val3)
2690 int cmd = op & FUTEX_CMD_MASK;
2691 unsigned int flags = 0;
2693 if (!(op & FUTEX_PRIVATE_FLAG))
2694 flags |= FLAGS_SHARED;
2696 if (op & FUTEX_CLOCK_REALTIME) {
2697 flags |= FLAGS_CLOCKRT;
2698 if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI)
2704 case FUTEX_UNLOCK_PI:
2705 case FUTEX_TRYLOCK_PI:
2706 case FUTEX_WAIT_REQUEUE_PI:
2707 case FUTEX_CMP_REQUEUE_PI:
2708 if (!futex_cmpxchg_enabled)
2714 val3 = FUTEX_BITSET_MATCH_ANY;
2715 case FUTEX_WAIT_BITSET:
2716 return futex_wait(uaddr, flags, val, timeout, val3);
2718 val3 = FUTEX_BITSET_MATCH_ANY;
2719 case FUTEX_WAKE_BITSET:
2720 return futex_wake(uaddr, flags, val, val3);
2722 return futex_requeue(uaddr, flags, uaddr2, val, val2, NULL, 0);
2723 case FUTEX_CMP_REQUEUE:
2724 return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 0);
2726 return futex_wake_op(uaddr, flags, uaddr2, val, val2, val3);
2728 return futex_lock_pi(uaddr, flags, val, timeout, 0);
2729 case FUTEX_UNLOCK_PI:
2730 return futex_unlock_pi(uaddr, flags);
2731 case FUTEX_TRYLOCK_PI:
2732 return futex_lock_pi(uaddr, flags, 0, timeout, 1);
2733 case FUTEX_WAIT_REQUEUE_PI:
2734 val3 = FUTEX_BITSET_MATCH_ANY;
2735 return futex_wait_requeue_pi(uaddr, flags, val, timeout, val3,
2737 case FUTEX_CMP_REQUEUE_PI:
2738 return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 1);
2744 SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val,
2745 struct timespec __user *, utime, u32 __user *, uaddr2,
2749 ktime_t t, *tp = NULL;
2751 int cmd = op & FUTEX_CMD_MASK;
2753 if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI ||
2754 cmd == FUTEX_WAIT_BITSET ||
2755 cmd == FUTEX_WAIT_REQUEUE_PI)) {
2756 if (copy_from_user(&ts, utime, sizeof(ts)) != 0)
2758 if (!timespec_valid(&ts))
2761 t = timespec_to_ktime(ts);
2762 if (cmd == FUTEX_WAIT)
2763 t = ktime_add_safe(ktime_get(), t);
2767 * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*.
2768 * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP.
2770 if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE ||
2771 cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP)
2772 val2 = (u32) (unsigned long) utime;
2774 return do_futex(uaddr, op, val, tp, uaddr2, val2, val3);
2777 static int __init futex_init(void)
2782 #if CONFIG_BASE_SMALL
2783 futex_hashsize = 16;
2785 futex_hashsize = roundup_pow_of_two(256 * num_possible_cpus());
2788 futex_queues = alloc_large_system_hash("futex", sizeof(*futex_queues),
2790 futex_hashsize < 256 ? HASH_SMALL : 0,
2791 NULL, NULL, futex_hashsize, futex_hashsize);
2794 * This will fail and we want it. Some arch implementations do
2795 * runtime detection of the futex_atomic_cmpxchg_inatomic()
2796 * functionality. We want to know that before we call in any
2797 * of the complex code paths. Also we want to prevent
2798 * registration of robust lists in that case. NULL is
2799 * guaranteed to fault and we get -EFAULT on functional
2800 * implementation, the non-functional ones will return
2803 if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT)
2804 futex_cmpxchg_enabled = 1;
2806 for (i = 0; i < futex_hashsize; i++) {
2807 plist_head_init(&futex_queues[i].chain);
2808 spin_lock_init(&futex_queues[i].lock);
2813 __initcall(futex_init);