2 * Fast Userspace Mutexes (which I call "Futexes!").
3 * (C) Rusty Russell, IBM 2002
5 * Generalized futexes, futex requeueing, misc fixes by Ingo Molnar
6 * (C) Copyright 2003 Red Hat Inc, All Rights Reserved
8 * Removed page pinning, fix privately mapped COW pages and other cleanups
9 * (C) Copyright 2003, 2004 Jamie Lokier
11 * Robust futex support started by Ingo Molnar
12 * (C) Copyright 2006 Red Hat Inc, All Rights Reserved
13 * Thanks to Thomas Gleixner for suggestions, analysis and fixes.
15 * PI-futex support started by Ingo Molnar and Thomas Gleixner
16 * Copyright (C) 2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
17 * Copyright (C) 2006 Timesys Corp., Thomas Gleixner <tglx@timesys.com>
19 * PRIVATE futexes by Eric Dumazet
20 * Copyright (C) 2007 Eric Dumazet <dada1@cosmosbay.com>
22 * Requeue-PI support by Darren Hart <dvhltc@us.ibm.com>
23 * Copyright (C) IBM Corporation, 2009
24 * Thanks to Thomas Gleixner for conceptual design and careful reviews.
26 * Thanks to Ben LaHaise for yelling "hashed waitqueues" loudly
27 * enough at me, Linus for the original (flawed) idea, Matthew
28 * Kirkwood for proof-of-concept implementation.
30 * "The futexes are also cursed."
31 * "But they come in a choice of three flavours!"
33 * This program is free software; you can redistribute it and/or modify
34 * it under the terms of the GNU General Public License as published by
35 * the Free Software Foundation; either version 2 of the License, or
36 * (at your option) any later version.
38 * This program is distributed in the hope that it will be useful,
39 * but WITHOUT ANY WARRANTY; without even the implied warranty of
40 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
41 * GNU General Public License for more details.
43 * You should have received a copy of the GNU General Public License
44 * along with this program; if not, write to the Free Software
45 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
47 #include <linux/slab.h>
48 #include <linux/poll.h>
50 #include <linux/file.h>
51 #include <linux/jhash.h>
52 #include <linux/init.h>
53 #include <linux/futex.h>
54 #include <linux/mount.h>
55 #include <linux/pagemap.h>
56 #include <linux/syscalls.h>
57 #include <linux/signal.h>
58 #include <linux/export.h>
59 #include <linux/magic.h>
60 #include <linux/pid.h>
61 #include <linux/nsproxy.h>
62 #include <linux/ptrace.h>
63 #include <linux/hugetlb.h>
65 #include <asm/futex.h>
67 #include "rtmutex_common.h"
69 int __read_mostly futex_cmpxchg_enabled;
71 #define FUTEX_HASHBITS (CONFIG_BASE_SMALL ? 4 : 8)
74 * Futex flags used to encode options to functions and preserve them across
77 #define FLAGS_SHARED 0x01
78 #define FLAGS_CLOCKRT 0x02
79 #define FLAGS_HAS_TIMEOUT 0x04
82 * Priority Inheritance state:
84 struct futex_pi_state {
86 * list of 'owned' pi_state instances - these have to be
87 * cleaned up in do_exit() if the task exits prematurely:
89 struct list_head list;
94 struct rt_mutex pi_mutex;
96 struct task_struct *owner;
103 * struct futex_q - The hashed futex queue entry, one per waiting task
104 * @list: priority-sorted list of tasks waiting on this futex
105 * @task: the task waiting on the futex
106 * @lock_ptr: the hash bucket lock
107 * @key: the key the futex is hashed on
108 * @pi_state: optional priority inheritance state
109 * @rt_waiter: rt_waiter storage for use with requeue_pi
110 * @requeue_pi_key: the requeue_pi target futex key
111 * @bitset: bitset for the optional bitmasked wakeup
113 * We use this hashed waitqueue, instead of a normal wait_queue_t, so
114 * we can wake only the relevant ones (hashed queues may be shared).
116 * A futex_q has a woken state, just like tasks have TASK_RUNNING.
117 * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0.
118 * The order of wakeup is always to make the first condition true, then
121 * PI futexes are typically woken before they are removed from the hash list via
122 * the rt_mutex code. See unqueue_me_pi().
125 struct plist_node list;
127 struct task_struct *task;
128 spinlock_t *lock_ptr;
130 struct futex_pi_state *pi_state;
131 struct rt_mutex_waiter *rt_waiter;
132 union futex_key *requeue_pi_key;
136 static const struct futex_q futex_q_init = {
137 /* list gets initialized in queue_me()*/
138 .key = FUTEX_KEY_INIT,
139 .bitset = FUTEX_BITSET_MATCH_ANY
143 * Hash buckets are shared by all the futex_keys that hash to the same
144 * location. Each key may have multiple futex_q structures, one for each task
145 * waiting on a futex.
147 struct futex_hash_bucket {
149 struct plist_head chain;
152 static struct futex_hash_bucket futex_queues[1<<FUTEX_HASHBITS];
155 * We hash on the keys returned from get_futex_key (see below).
157 static struct futex_hash_bucket *hash_futex(union futex_key *key)
159 u32 hash = jhash2((u32*)&key->both.word,
160 (sizeof(key->both.word)+sizeof(key->both.ptr))/4,
162 return &futex_queues[hash & ((1 << FUTEX_HASHBITS)-1)];
166 * Return 1 if two futex_keys are equal, 0 otherwise.
168 static inline int match_futex(union futex_key *key1, union futex_key *key2)
171 && key1->both.word == key2->both.word
172 && key1->both.ptr == key2->both.ptr
173 && key1->both.offset == key2->both.offset);
177 * Take a reference to the resource addressed by a key.
178 * Can be called while holding spinlocks.
181 static void get_futex_key_refs(union futex_key *key)
186 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
188 ihold(key->shared.inode);
190 case FUT_OFF_MMSHARED:
191 atomic_inc(&key->private.mm->mm_count);
197 * Drop a reference to the resource addressed by a key.
198 * The hash bucket spinlock must not be held.
200 static void drop_futex_key_refs(union futex_key *key)
202 if (!key->both.ptr) {
203 /* If we're here then we tried to put a key we failed to get */
208 switch (key->both.offset & (FUT_OFF_INODE|FUT_OFF_MMSHARED)) {
210 iput(key->shared.inode);
212 case FUT_OFF_MMSHARED:
213 mmdrop(key->private.mm);
219 * get_futex_key() - Get parameters which are the keys for a futex
220 * @uaddr: virtual address of the futex
221 * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED
222 * @key: address where result is stored.
223 * @rw: mapping needs to be read/write (values: VERIFY_READ,
226 * Returns a negative error code or 0
227 * The key words are stored in *key on success.
229 * For shared mappings, it's (page->index, vma->vm_file->f_path.dentry->d_inode,
230 * offset_within_page). For private mappings, it's (uaddr, current->mm).
231 * We can usually work out the index without swapping in the page.
233 * lock_page() might sleep, the caller should not hold a spinlock.
236 get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key, int rw)
238 unsigned long address = (unsigned long)uaddr;
239 struct mm_struct *mm = current->mm;
240 struct page *page, *page_head;
244 * The futex address must be "naturally" aligned.
246 key->both.offset = address % PAGE_SIZE;
247 if (unlikely((address % sizeof(u32)) != 0))
249 address -= key->both.offset;
252 * PROCESS_PRIVATE futexes are fast.
253 * As the mm cannot disappear under us and the 'key' only needs
254 * virtual address, we dont even have to find the underlying vma.
255 * Note : We do have to check 'uaddr' is a valid user address,
256 * but access_ok() should be faster than find_vma()
259 if (unlikely(!access_ok(VERIFY_WRITE, uaddr, sizeof(u32))))
261 key->private.mm = mm;
262 key->private.address = address;
263 get_futex_key_refs(key);
268 err = get_user_pages_fast(address, 1, 1, &page);
270 * If write access is not required (eg. FUTEX_WAIT), try
271 * and get read-only access.
273 if (err == -EFAULT && rw == VERIFY_READ) {
274 err = get_user_pages_fast(address, 1, 0, &page);
282 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
284 if (unlikely(PageTail(page))) {
286 /* serialize against __split_huge_page_splitting() */
288 if (likely(__get_user_pages_fast(address, 1, !ro, &page) == 1)) {
289 page_head = compound_head(page);
291 * page_head is valid pointer but we must pin
292 * it before taking the PG_lock and/or
293 * PG_compound_lock. The moment we re-enable
294 * irqs __split_huge_page_splitting() can
295 * return and the head page can be freed from
296 * under us. We can't take the PG_lock and/or
297 * PG_compound_lock on a page that could be
298 * freed from under us.
300 if (page != page_head) {
311 page_head = compound_head(page);
312 if (page != page_head) {
318 lock_page(page_head);
321 * If page_head->mapping is NULL, then it cannot be a PageAnon
322 * page; but it might be the ZERO_PAGE or in the gate area or
323 * in a special mapping (all cases which we are happy to fail);
324 * or it may have been a good file page when get_user_pages_fast
325 * found it, but truncated or holepunched or subjected to
326 * invalidate_complete_page2 before we got the page lock (also
327 * cases which we are happy to fail). And we hold a reference,
328 * so refcount care in invalidate_complete_page's remove_mapping
329 * prevents drop_caches from setting mapping to NULL beneath us.
331 * The case we do have to guard against is when memory pressure made
332 * shmem_writepage move it from filecache to swapcache beneath us:
333 * an unlikely race, but we do need to retry for page_head->mapping.
335 if (!page_head->mapping) {
336 int shmem_swizzled = PageSwapCache(page_head);
337 unlock_page(page_head);
345 * Private mappings are handled in a simple way.
347 * NOTE: When userspace waits on a MAP_SHARED mapping, even if
348 * it's a read-only handle, it's expected that futexes attach to
349 * the object not the particular process.
351 if (PageAnon(page_head)) {
353 * A RO anonymous page will never change and thus doesn't make
354 * sense for futex operations.
361 key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */
362 key->private.mm = mm;
363 key->private.address = address;
365 key->both.offset |= FUT_OFF_INODE; /* inode-based key */
366 key->shared.inode = page_head->mapping->host;
367 key->shared.pgoff = basepage_index(page);
370 get_futex_key_refs(key);
373 unlock_page(page_head);
378 static inline void put_futex_key(union futex_key *key)
380 drop_futex_key_refs(key);
384 * fault_in_user_writeable() - Fault in user address and verify RW access
385 * @uaddr: pointer to faulting user space address
387 * Slow path to fixup the fault we just took in the atomic write
390 * We have no generic implementation of a non-destructive write to the
391 * user address. We know that we faulted in the atomic pagefault
392 * disabled section so we can as well avoid the #PF overhead by
393 * calling get_user_pages() right away.
395 static int fault_in_user_writeable(u32 __user *uaddr)
397 struct mm_struct *mm = current->mm;
400 down_read(&mm->mmap_sem);
401 ret = fixup_user_fault(current, mm, (unsigned long)uaddr,
403 up_read(&mm->mmap_sem);
405 return ret < 0 ? ret : 0;
409 * futex_top_waiter() - Return the highest priority waiter on a futex
410 * @hb: the hash bucket the futex_q's reside in
411 * @key: the futex key (to distinguish it from other futex futex_q's)
413 * Must be called with the hb lock held.
415 static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb,
416 union futex_key *key)
418 struct futex_q *this;
420 plist_for_each_entry(this, &hb->chain, list) {
421 if (match_futex(&this->key, key))
427 static int cmpxchg_futex_value_locked(u32 *curval, u32 __user *uaddr,
428 u32 uval, u32 newval)
433 ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval);
439 static int get_futex_value_locked(u32 *dest, u32 __user *from)
444 ret = __copy_from_user_inatomic(dest, from, sizeof(u32));
447 return ret ? -EFAULT : 0;
454 static int refill_pi_state_cache(void)
456 struct futex_pi_state *pi_state;
458 if (likely(current->pi_state_cache))
461 pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL);
466 INIT_LIST_HEAD(&pi_state->list);
467 /* pi_mutex gets initialized later */
468 pi_state->owner = NULL;
469 atomic_set(&pi_state->refcount, 1);
470 pi_state->key = FUTEX_KEY_INIT;
472 current->pi_state_cache = pi_state;
477 static struct futex_pi_state * alloc_pi_state(void)
479 struct futex_pi_state *pi_state = current->pi_state_cache;
482 current->pi_state_cache = NULL;
488 * Must be called with the hb lock held.
490 static void free_pi_state(struct futex_pi_state *pi_state)
495 if (!atomic_dec_and_test(&pi_state->refcount))
499 * If pi_state->owner is NULL, the owner is most probably dying
500 * and has cleaned up the pi_state already
502 if (pi_state->owner) {
503 raw_spin_lock_irq(&pi_state->owner->pi_lock);
504 list_del_init(&pi_state->list);
505 raw_spin_unlock_irq(&pi_state->owner->pi_lock);
507 rt_mutex_proxy_unlock(&pi_state->pi_mutex, pi_state->owner);
510 if (current->pi_state_cache)
514 * pi_state->list is already empty.
515 * clear pi_state->owner.
516 * refcount is at 0 - put it back to 1.
518 pi_state->owner = NULL;
519 atomic_set(&pi_state->refcount, 1);
520 current->pi_state_cache = pi_state;
525 * Look up the task based on what TID userspace gave us.
528 static struct task_struct * futex_find_get_task(pid_t pid)
530 struct task_struct *p;
533 p = find_task_by_vpid(pid);
543 * This task is holding PI mutexes at exit time => bad.
544 * Kernel cleans up PI-state, but userspace is likely hosed.
545 * (Robust-futex cleanup is separate and might save the day for userspace.)
547 void exit_pi_state_list(struct task_struct *curr)
549 struct list_head *next, *head = &curr->pi_state_list;
550 struct futex_pi_state *pi_state;
551 struct futex_hash_bucket *hb;
552 union futex_key key = FUTEX_KEY_INIT;
554 if (!futex_cmpxchg_enabled)
557 * We are a ZOMBIE and nobody can enqueue itself on
558 * pi_state_list anymore, but we have to be careful
559 * versus waiters unqueueing themselves:
561 raw_spin_lock_irq(&curr->pi_lock);
562 while (!list_empty(head)) {
565 pi_state = list_entry(next, struct futex_pi_state, list);
567 hb = hash_futex(&key);
568 raw_spin_unlock_irq(&curr->pi_lock);
570 spin_lock(&hb->lock);
572 raw_spin_lock_irq(&curr->pi_lock);
574 * We dropped the pi-lock, so re-check whether this
575 * task still owns the PI-state:
577 if (head->next != next) {
578 spin_unlock(&hb->lock);
582 WARN_ON(pi_state->owner != curr);
583 WARN_ON(list_empty(&pi_state->list));
584 list_del_init(&pi_state->list);
585 pi_state->owner = NULL;
586 raw_spin_unlock_irq(&curr->pi_lock);
588 rt_mutex_unlock(&pi_state->pi_mutex);
590 spin_unlock(&hb->lock);
592 raw_spin_lock_irq(&curr->pi_lock);
594 raw_spin_unlock_irq(&curr->pi_lock);
598 * We need to check the following states:
600 * Waiter | pi_state | pi->owner | uTID | uODIED | ?
602 * [1] NULL | --- | --- | 0 | 0/1 | Valid
603 * [2] NULL | --- | --- | >0 | 0/1 | Valid
605 * [3] Found | NULL | -- | Any | 0/1 | Invalid
607 * [4] Found | Found | NULL | 0 | 1 | Valid
608 * [5] Found | Found | NULL | >0 | 1 | Invalid
610 * [6] Found | Found | task | 0 | 1 | Valid
612 * [7] Found | Found | NULL | Any | 0 | Invalid
614 * [8] Found | Found | task | ==taskTID | 0/1 | Valid
615 * [9] Found | Found | task | 0 | 0 | Invalid
616 * [10] Found | Found | task | !=taskTID | 0/1 | Invalid
618 * [1] Indicates that the kernel can acquire the futex atomically. We
619 * came came here due to a stale FUTEX_WAITERS/FUTEX_OWNER_DIED bit.
621 * [2] Valid, if TID does not belong to a kernel thread. If no matching
622 * thread is found then it indicates that the owner TID has died.
624 * [3] Invalid. The waiter is queued on a non PI futex
626 * [4] Valid state after exit_robust_list(), which sets the user space
627 * value to FUTEX_WAITERS | FUTEX_OWNER_DIED.
629 * [5] The user space value got manipulated between exit_robust_list()
630 * and exit_pi_state_list()
632 * [6] Valid state after exit_pi_state_list() which sets the new owner in
633 * the pi_state but cannot access the user space value.
635 * [7] pi_state->owner can only be NULL when the OWNER_DIED bit is set.
637 * [8] Owner and user space value match
639 * [9] There is no transient state which sets the user space TID to 0
640 * except exit_robust_list(), but this is indicated by the
641 * FUTEX_OWNER_DIED bit. See [4]
643 * [10] There is no transient state which leaves owner and user space
647 lookup_pi_state(u32 uval, struct futex_hash_bucket *hb,
648 union futex_key *key, struct futex_pi_state **ps)
650 struct futex_pi_state *pi_state = NULL;
651 struct futex_q *this, *next;
652 struct plist_head *head;
653 struct task_struct *p;
654 pid_t pid = uval & FUTEX_TID_MASK;
658 plist_for_each_entry_safe(this, next, head, list) {
659 if (match_futex(&this->key, key)) {
661 * Sanity check the waiter before increasing
662 * the refcount and attaching to it.
664 pi_state = this->pi_state;
666 * Userspace might have messed up non-PI and
669 if (unlikely(!pi_state))
672 WARN_ON(!atomic_read(&pi_state->refcount));
675 * Handle the owner died case:
677 if (uval & FUTEX_OWNER_DIED) {
679 * exit_pi_state_list sets owner to NULL and
680 * wakes the topmost waiter. The task which
681 * acquires the pi_state->rt_mutex will fixup
684 if (!pi_state->owner) {
686 * No pi state owner, but the user
687 * space TID is not 0. Inconsistent
693 * Take a ref on the state and
700 * If TID is 0, then either the dying owner
701 * has not yet executed exit_pi_state_list()
702 * or some waiter acquired the rtmutex in the
703 * pi state, but did not yet fixup the TID in
706 * Take a ref on the state and return. [6]
712 * If the owner died bit is not set,
713 * then the pi_state must have an
716 if (!pi_state->owner)
721 * Bail out if user space manipulated the
722 * futex value. If pi state exists then the
723 * owner TID must be the same as the user
726 if (pid != task_pid_vnr(pi_state->owner))
730 atomic_inc(&pi_state->refcount);
737 * We are the first waiter - try to look up the real owner and attach
738 * the new pi_state to it, but bail out when TID = 0 [1]
742 p = futex_find_get_task(pid);
752 * We need to look at the task state flags to figure out,
753 * whether the task is exiting. To protect against the do_exit
754 * change of the task flags, we do this protected by
757 raw_spin_lock_irq(&p->pi_lock);
758 if (unlikely(p->flags & PF_EXITING)) {
760 * The task is on the way out. When PF_EXITPIDONE is
761 * set, we know that the task has finished the
764 int ret = (p->flags & PF_EXITPIDONE) ? -ESRCH : -EAGAIN;
766 raw_spin_unlock_irq(&p->pi_lock);
772 * No existing pi state. First waiter. [2]
774 pi_state = alloc_pi_state();
777 * Initialize the pi_mutex in locked state and make 'p'
780 rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p);
782 /* Store the key for possible exit cleanups: */
783 pi_state->key = *key;
785 WARN_ON(!list_empty(&pi_state->list));
786 list_add(&pi_state->list, &p->pi_state_list);
788 raw_spin_unlock_irq(&p->pi_lock);
798 * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex
799 * @uaddr: the pi futex user address
800 * @hb: the pi futex hash bucket
801 * @key: the futex key associated with uaddr and hb
802 * @ps: the pi_state pointer where we store the result of the
804 * @task: the task to perform the atomic lock work for. This will
805 * be "current" except in the case of requeue pi.
806 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
810 * 1 - acquired the lock
813 * The hb->lock and futex_key refs shall be held by the caller.
815 static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb,
816 union futex_key *key,
817 struct futex_pi_state **ps,
818 struct task_struct *task, int set_waiters)
820 int lock_taken, ret, force_take = 0;
821 u32 uval, newval, curval, vpid = task_pid_vnr(task);
824 ret = lock_taken = 0;
827 * To avoid races, we attempt to take the lock here again
828 * (by doing a 0 -> TID atomic cmpxchg), while holding all
829 * the locks. It will most likely not succeed.
833 newval |= FUTEX_WAITERS;
835 if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, 0, newval)))
841 if ((unlikely((curval & FUTEX_TID_MASK) == vpid)))
845 * Surprise - we got the lock, but we do not trust user space at all.
847 if (unlikely(!curval)) {
849 * We verify whether there is kernel state for this
850 * futex. If not, we can safely assume, that the 0 ->
851 * TID transition is correct. If state exists, we do
852 * not bother to fixup the user space state as it was
855 return futex_top_waiter(hb, key) ? -EINVAL : 1;
861 * Set the FUTEX_WAITERS flag, so the owner will know it has someone
862 * to wake at the next unlock.
864 newval = curval | FUTEX_WAITERS;
867 * Should we force take the futex? See below.
869 if (unlikely(force_take)) {
871 * Keep the OWNER_DIED and the WAITERS bit and set the
874 newval = (curval & ~FUTEX_TID_MASK) | vpid;
879 if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)))
881 if (unlikely(curval != uval))
885 * We took the lock due to forced take over.
887 if (unlikely(lock_taken))
891 * We dont have the lock. Look up the PI state (or create it if
892 * we are the first waiter):
894 ret = lookup_pi_state(uval, hb, key, ps);
900 * We failed to find an owner for this
901 * futex. So we have no pi_state to block
902 * on. This can happen in two cases:
905 * 2) A stale FUTEX_WAITERS bit
907 * Re-read the futex value.
909 if (get_futex_value_locked(&curval, uaddr))
913 * If the owner died or we have a stale
914 * WAITERS bit the owner TID in the user space
917 if (!(curval & FUTEX_TID_MASK)) {
930 * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket
931 * @q: The futex_q to unqueue
933 * The q->lock_ptr must not be NULL and must be held by the caller.
935 static void __unqueue_futex(struct futex_q *q)
937 struct futex_hash_bucket *hb;
939 if (WARN_ON_SMP(!q->lock_ptr || !spin_is_locked(q->lock_ptr))
940 || WARN_ON(plist_node_empty(&q->list)))
943 hb = container_of(q->lock_ptr, struct futex_hash_bucket, lock);
944 plist_del(&q->list, &hb->chain);
948 * The hash bucket lock must be held when this is called.
949 * Afterwards, the futex_q must not be accessed.
951 static void wake_futex(struct futex_q *q)
953 struct task_struct *p = q->task;
955 if (WARN(q->pi_state || q->rt_waiter, "refusing to wake PI futex\n"))
959 * We set q->lock_ptr = NULL _before_ we wake up the task. If
960 * a non-futex wake up happens on another CPU then the task
961 * might exit and p would dereference a non-existing task
962 * struct. Prevent this by holding a reference on p across the
969 * The waiting task can free the futex_q as soon as
970 * q->lock_ptr = NULL is written, without taking any locks. A
971 * memory barrier is required here to prevent the following
972 * store to lock_ptr from getting ahead of the plist_del.
977 wake_up_state(p, TASK_NORMAL);
981 static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this)
983 struct task_struct *new_owner;
984 struct futex_pi_state *pi_state = this->pi_state;
985 u32 uninitialized_var(curval), newval;
992 * If current does not own the pi_state then the futex is
993 * inconsistent and user space fiddled with the futex value.
995 if (pi_state->owner != current)
998 raw_spin_lock(&pi_state->pi_mutex.wait_lock);
999 new_owner = rt_mutex_next_owner(&pi_state->pi_mutex);
1002 * It is possible that the next waiter (the one that brought
1003 * this owner to the kernel) timed out and is no longer
1004 * waiting on the lock.
1007 new_owner = this->task;
1010 * We pass it to the next owner. The WAITERS bit is always
1011 * kept enabled while there is PI state around. We cleanup the
1012 * owner died bit, because we are the owner.
1014 newval = FUTEX_WAITERS | task_pid_vnr(new_owner);
1016 if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))
1018 else if (curval != uval)
1021 raw_spin_unlock(&pi_state->pi_mutex.wait_lock);
1025 raw_spin_lock_irq(&pi_state->owner->pi_lock);
1026 WARN_ON(list_empty(&pi_state->list));
1027 list_del_init(&pi_state->list);
1028 raw_spin_unlock_irq(&pi_state->owner->pi_lock);
1030 raw_spin_lock_irq(&new_owner->pi_lock);
1031 WARN_ON(!list_empty(&pi_state->list));
1032 list_add(&pi_state->list, &new_owner->pi_state_list);
1033 pi_state->owner = new_owner;
1034 raw_spin_unlock_irq(&new_owner->pi_lock);
1036 raw_spin_unlock(&pi_state->pi_mutex.wait_lock);
1037 rt_mutex_unlock(&pi_state->pi_mutex);
1042 static int unlock_futex_pi(u32 __user *uaddr, u32 uval)
1044 u32 uninitialized_var(oldval);
1047 * There is no waiter, so we unlock the futex. The owner died
1048 * bit has not to be preserved here. We are the owner:
1050 if (cmpxchg_futex_value_locked(&oldval, uaddr, uval, 0))
1059 * Express the locking dependencies for lockdep:
1062 double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
1065 spin_lock(&hb1->lock);
1067 spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING);
1068 } else { /* hb1 > hb2 */
1069 spin_lock(&hb2->lock);
1070 spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING);
1075 double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
1077 spin_unlock(&hb1->lock);
1079 spin_unlock(&hb2->lock);
1083 * Wake up waiters matching bitset queued on this futex (uaddr).
1086 futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset)
1088 struct futex_hash_bucket *hb;
1089 struct futex_q *this, *next;
1090 struct plist_head *head;
1091 union futex_key key = FUTEX_KEY_INIT;
1097 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_READ);
1098 if (unlikely(ret != 0))
1101 hb = hash_futex(&key);
1102 spin_lock(&hb->lock);
1105 plist_for_each_entry_safe(this, next, head, list) {
1106 if (match_futex (&this->key, &key)) {
1107 if (this->pi_state || this->rt_waiter) {
1112 /* Check if one of the bits is set in both bitsets */
1113 if (!(this->bitset & bitset))
1117 if (++ret >= nr_wake)
1122 spin_unlock(&hb->lock);
1123 put_futex_key(&key);
1129 * Wake up all waiters hashed on the physical page that is mapped
1130 * to this virtual address:
1133 futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2,
1134 int nr_wake, int nr_wake2, int op)
1136 union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
1137 struct futex_hash_bucket *hb1, *hb2;
1138 struct plist_head *head;
1139 struct futex_q *this, *next;
1143 ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ);
1144 if (unlikely(ret != 0))
1146 ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE);
1147 if (unlikely(ret != 0))
1150 hb1 = hash_futex(&key1);
1151 hb2 = hash_futex(&key2);
1154 double_lock_hb(hb1, hb2);
1155 op_ret = futex_atomic_op_inuser(op, uaddr2);
1156 if (unlikely(op_ret < 0)) {
1158 double_unlock_hb(hb1, hb2);
1162 * we don't get EFAULT from MMU faults if we don't have an MMU,
1163 * but we might get them from range checking
1169 if (unlikely(op_ret != -EFAULT)) {
1174 ret = fault_in_user_writeable(uaddr2);
1178 if (!(flags & FLAGS_SHARED))
1181 put_futex_key(&key2);
1182 put_futex_key(&key1);
1188 plist_for_each_entry_safe(this, next, head, list) {
1189 if (match_futex (&this->key, &key1)) {
1190 if (this->pi_state || this->rt_waiter) {
1195 if (++ret >= nr_wake)
1204 plist_for_each_entry_safe(this, next, head, list) {
1205 if (match_futex (&this->key, &key2)) {
1206 if (this->pi_state || this->rt_waiter) {
1211 if (++op_ret >= nr_wake2)
1219 double_unlock_hb(hb1, hb2);
1221 put_futex_key(&key2);
1223 put_futex_key(&key1);
1229 * requeue_futex() - Requeue a futex_q from one hb to another
1230 * @q: the futex_q to requeue
1231 * @hb1: the source hash_bucket
1232 * @hb2: the target hash_bucket
1233 * @key2: the new key for the requeued futex_q
1236 void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1,
1237 struct futex_hash_bucket *hb2, union futex_key *key2)
1241 * If key1 and key2 hash to the same bucket, no need to
1244 if (likely(&hb1->chain != &hb2->chain)) {
1245 plist_del(&q->list, &hb1->chain);
1246 plist_add(&q->list, &hb2->chain);
1247 q->lock_ptr = &hb2->lock;
1249 get_futex_key_refs(key2);
1254 * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue
1256 * @key: the key of the requeue target futex
1257 * @hb: the hash_bucket of the requeue target futex
1259 * During futex_requeue, with requeue_pi=1, it is possible to acquire the
1260 * target futex if it is uncontended or via a lock steal. Set the futex_q key
1261 * to the requeue target futex so the waiter can detect the wakeup on the right
1262 * futex, but remove it from the hb and NULL the rt_waiter so it can detect
1263 * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock
1264 * to protect access to the pi_state to fixup the owner later. Must be called
1265 * with both q->lock_ptr and hb->lock held.
1268 void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key,
1269 struct futex_hash_bucket *hb)
1271 get_futex_key_refs(key);
1276 WARN_ON(!q->rt_waiter);
1277 q->rt_waiter = NULL;
1279 q->lock_ptr = &hb->lock;
1281 wake_up_state(q->task, TASK_NORMAL);
1285 * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter
1286 * @pifutex: the user address of the to futex
1287 * @hb1: the from futex hash bucket, must be locked by the caller
1288 * @hb2: the to futex hash bucket, must be locked by the caller
1289 * @key1: the from futex key
1290 * @key2: the to futex key
1291 * @ps: address to store the pi_state pointer
1292 * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
1294 * Try and get the lock on behalf of the top waiter if we can do it atomically.
1295 * Wake the top waiter if we succeed. If the caller specified set_waiters,
1296 * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit.
1297 * hb1 and hb2 must be held by the caller.
1300 * 0 - failed to acquire the lock atomicly
1301 * >0 - acquired the lock, return value is vpid of the top_waiter
1304 static int futex_proxy_trylock_atomic(u32 __user *pifutex,
1305 struct futex_hash_bucket *hb1,
1306 struct futex_hash_bucket *hb2,
1307 union futex_key *key1, union futex_key *key2,
1308 struct futex_pi_state **ps, int set_waiters)
1310 struct futex_q *top_waiter = NULL;
1314 if (get_futex_value_locked(&curval, pifutex))
1318 * Find the top_waiter and determine if there are additional waiters.
1319 * If the caller intends to requeue more than 1 waiter to pifutex,
1320 * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now,
1321 * as we have means to handle the possible fault. If not, don't set
1322 * the bit unecessarily as it will force the subsequent unlock to enter
1325 top_waiter = futex_top_waiter(hb1, key1);
1327 /* There are no waiters, nothing for us to do. */
1331 /* Ensure we requeue to the expected futex. */
1332 if (!match_futex(top_waiter->requeue_pi_key, key2))
1336 * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in
1337 * the contended case or if set_waiters is 1. The pi_state is returned
1338 * in ps in contended cases.
1340 vpid = task_pid_vnr(top_waiter->task);
1341 ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task,
1344 requeue_pi_wake_futex(top_waiter, key2, hb2);
1351 * futex_requeue() - Requeue waiters from uaddr1 to uaddr2
1352 * @uaddr1: source futex user address
1353 * @flags: futex flags (FLAGS_SHARED, etc.)
1354 * @uaddr2: target futex user address
1355 * @nr_wake: number of waiters to wake (must be 1 for requeue_pi)
1356 * @nr_requeue: number of waiters to requeue (0-INT_MAX)
1357 * @cmpval: @uaddr1 expected value (or %NULL)
1358 * @requeue_pi: if we are attempting to requeue from a non-pi futex to a
1359 * pi futex (pi to pi requeue is not supported)
1361 * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire
1362 * uaddr2 atomically on behalf of the top waiter.
1365 * >=0 - on success, the number of tasks requeued or woken
1368 static int futex_requeue(u32 __user *uaddr1, unsigned int flags,
1369 u32 __user *uaddr2, int nr_wake, int nr_requeue,
1370 u32 *cmpval, int requeue_pi)
1372 union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
1373 int drop_count = 0, task_count = 0, ret;
1374 struct futex_pi_state *pi_state = NULL;
1375 struct futex_hash_bucket *hb1, *hb2;
1376 struct plist_head *head1;
1377 struct futex_q *this, *next;
1381 * Requeue PI only works on two distinct uaddrs. This
1382 * check is only valid for private futexes. See below.
1384 if (uaddr1 == uaddr2)
1388 * requeue_pi requires a pi_state, try to allocate it now
1389 * without any locks in case it fails.
1391 if (refill_pi_state_cache())
1394 * requeue_pi must wake as many tasks as it can, up to nr_wake
1395 * + nr_requeue, since it acquires the rt_mutex prior to
1396 * returning to userspace, so as to not leave the rt_mutex with
1397 * waiters and no owner. However, second and third wake-ups
1398 * cannot be predicted as they involve race conditions with the
1399 * first wake and a fault while looking up the pi_state. Both
1400 * pthread_cond_signal() and pthread_cond_broadcast() should
1408 ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, VERIFY_READ);
1409 if (unlikely(ret != 0))
1411 ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2,
1412 requeue_pi ? VERIFY_WRITE : VERIFY_READ);
1413 if (unlikely(ret != 0))
1417 * The check above which compares uaddrs is not sufficient for
1418 * shared futexes. We need to compare the keys:
1420 if (requeue_pi && match_futex(&key1, &key2)) {
1425 hb1 = hash_futex(&key1);
1426 hb2 = hash_futex(&key2);
1429 double_lock_hb(hb1, hb2);
1431 if (likely(cmpval != NULL)) {
1434 ret = get_futex_value_locked(&curval, uaddr1);
1436 if (unlikely(ret)) {
1437 double_unlock_hb(hb1, hb2);
1439 ret = get_user(curval, uaddr1);
1443 if (!(flags & FLAGS_SHARED))
1446 put_futex_key(&key2);
1447 put_futex_key(&key1);
1450 if (curval != *cmpval) {
1456 if (requeue_pi && (task_count - nr_wake < nr_requeue)) {
1458 * Attempt to acquire uaddr2 and wake the top waiter. If we
1459 * intend to requeue waiters, force setting the FUTEX_WAITERS
1460 * bit. We force this here where we are able to easily handle
1461 * faults rather in the requeue loop below.
1463 ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1,
1464 &key2, &pi_state, nr_requeue);
1467 * At this point the top_waiter has either taken uaddr2 or is
1468 * waiting on it. If the former, then the pi_state will not
1469 * exist yet, look it up one more time to ensure we have a
1470 * reference to it. If the lock was taken, ret contains the
1471 * vpid of the top waiter task.
1478 * If we acquired the lock, then the user
1479 * space value of uaddr2 should be vpid. It
1480 * cannot be changed by the top waiter as it
1481 * is blocked on hb2 lock if it tries to do
1482 * so. If something fiddled with it behind our
1483 * back the pi state lookup might unearth
1484 * it. So we rather use the known value than
1485 * rereading and handing potential crap to
1488 ret = lookup_pi_state(ret, hb2, &key2, &pi_state);
1495 free_pi_state(pi_state);
1497 double_unlock_hb(hb1, hb2);
1498 put_futex_key(&key2);
1499 put_futex_key(&key1);
1500 ret = fault_in_user_writeable(uaddr2);
1505 /* The owner was exiting, try again. */
1506 free_pi_state(pi_state);
1508 double_unlock_hb(hb1, hb2);
1509 put_futex_key(&key2);
1510 put_futex_key(&key1);
1518 head1 = &hb1->chain;
1519 plist_for_each_entry_safe(this, next, head1, list) {
1520 if (task_count - nr_wake >= nr_requeue)
1523 if (!match_futex(&this->key, &key1))
1527 * FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always
1528 * be paired with each other and no other futex ops.
1530 * We should never be requeueing a futex_q with a pi_state,
1531 * which is awaiting a futex_unlock_pi().
1533 if ((requeue_pi && !this->rt_waiter) ||
1534 (!requeue_pi && this->rt_waiter) ||
1541 * Wake nr_wake waiters. For requeue_pi, if we acquired the
1542 * lock, we already woke the top_waiter. If not, it will be
1543 * woken by futex_unlock_pi().
1545 if (++task_count <= nr_wake && !requeue_pi) {
1550 /* Ensure we requeue to the expected futex for requeue_pi. */
1551 if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) {
1557 * Requeue nr_requeue waiters and possibly one more in the case
1558 * of requeue_pi if we couldn't acquire the lock atomically.
1561 /* Prepare the waiter to take the rt_mutex. */
1562 atomic_inc(&pi_state->refcount);
1563 this->pi_state = pi_state;
1564 ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex,
1568 /* We got the lock. */
1569 requeue_pi_wake_futex(this, &key2, hb2);
1574 this->pi_state = NULL;
1575 free_pi_state(pi_state);
1579 requeue_futex(this, hb1, hb2, &key2);
1584 free_pi_state(pi_state);
1585 double_unlock_hb(hb1, hb2);
1588 * drop_futex_key_refs() must be called outside the spinlocks. During
1589 * the requeue we moved futex_q's from the hash bucket at key1 to the
1590 * one at key2 and updated their key pointer. We no longer need to
1591 * hold the references to key1.
1593 while (--drop_count >= 0)
1594 drop_futex_key_refs(&key1);
1597 put_futex_key(&key2);
1599 put_futex_key(&key1);
1601 return ret ? ret : task_count;
1604 /* The key must be already stored in q->key. */
1605 static inline struct futex_hash_bucket *queue_lock(struct futex_q *q)
1606 __acquires(&hb->lock)
1608 struct futex_hash_bucket *hb;
1610 hb = hash_futex(&q->key);
1611 q->lock_ptr = &hb->lock;
1613 spin_lock(&hb->lock);
1618 queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb)
1619 __releases(&hb->lock)
1621 spin_unlock(&hb->lock);
1625 * queue_me() - Enqueue the futex_q on the futex_hash_bucket
1626 * @q: The futex_q to enqueue
1627 * @hb: The destination hash bucket
1629 * The hb->lock must be held by the caller, and is released here. A call to
1630 * queue_me() is typically paired with exactly one call to unqueue_me(). The
1631 * exceptions involve the PI related operations, which may use unqueue_me_pi()
1632 * or nothing if the unqueue is done as part of the wake process and the unqueue
1633 * state is implicit in the state of woken task (see futex_wait_requeue_pi() for
1636 static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb)
1637 __releases(&hb->lock)
1642 * The priority used to register this element is
1643 * - either the real thread-priority for the real-time threads
1644 * (i.e. threads with a priority lower than MAX_RT_PRIO)
1645 * - or MAX_RT_PRIO for non-RT threads.
1646 * Thus, all RT-threads are woken first in priority order, and
1647 * the others are woken last, in FIFO order.
1649 prio = min(current->normal_prio, MAX_RT_PRIO);
1651 plist_node_init(&q->list, prio);
1652 plist_add(&q->list, &hb->chain);
1654 spin_unlock(&hb->lock);
1658 * unqueue_me() - Remove the futex_q from its futex_hash_bucket
1659 * @q: The futex_q to unqueue
1661 * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must
1662 * be paired with exactly one earlier call to queue_me().
1665 * 1 - if the futex_q was still queued (and we removed unqueued it)
1666 * 0 - if the futex_q was already removed by the waking thread
1668 static int unqueue_me(struct futex_q *q)
1670 spinlock_t *lock_ptr;
1673 /* In the common case we don't take the spinlock, which is nice. */
1675 lock_ptr = q->lock_ptr;
1677 if (lock_ptr != NULL) {
1678 spin_lock(lock_ptr);
1680 * q->lock_ptr can change between reading it and
1681 * spin_lock(), causing us to take the wrong lock. This
1682 * corrects the race condition.
1684 * Reasoning goes like this: if we have the wrong lock,
1685 * q->lock_ptr must have changed (maybe several times)
1686 * between reading it and the spin_lock(). It can
1687 * change again after the spin_lock() but only if it was
1688 * already changed before the spin_lock(). It cannot,
1689 * however, change back to the original value. Therefore
1690 * we can detect whether we acquired the correct lock.
1692 if (unlikely(lock_ptr != q->lock_ptr)) {
1693 spin_unlock(lock_ptr);
1698 BUG_ON(q->pi_state);
1700 spin_unlock(lock_ptr);
1704 drop_futex_key_refs(&q->key);
1709 * PI futexes can not be requeued and must remove themself from the
1710 * hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry
1713 static void unqueue_me_pi(struct futex_q *q)
1714 __releases(q->lock_ptr)
1718 BUG_ON(!q->pi_state);
1719 free_pi_state(q->pi_state);
1722 spin_unlock(q->lock_ptr);
1726 * Fixup the pi_state owner with the new owner.
1728 * Must be called with hash bucket lock held and mm->sem held for non
1731 static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q,
1732 struct task_struct *newowner)
1734 u32 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS;
1735 struct futex_pi_state *pi_state = q->pi_state;
1736 struct task_struct *oldowner = pi_state->owner;
1737 u32 uval, uninitialized_var(curval), newval;
1741 if (!pi_state->owner)
1742 newtid |= FUTEX_OWNER_DIED;
1745 * We are here either because we stole the rtmutex from the
1746 * previous highest priority waiter or we are the highest priority
1747 * waiter but failed to get the rtmutex the first time.
1748 * We have to replace the newowner TID in the user space variable.
1749 * This must be atomic as we have to preserve the owner died bit here.
1751 * Note: We write the user space value _before_ changing the pi_state
1752 * because we can fault here. Imagine swapped out pages or a fork
1753 * that marked all the anonymous memory readonly for cow.
1755 * Modifying pi_state _before_ the user space value would
1756 * leave the pi_state in an inconsistent state when we fault
1757 * here, because we need to drop the hash bucket lock to
1758 * handle the fault. This might be observed in the PID check
1759 * in lookup_pi_state.
1762 if (get_futex_value_locked(&uval, uaddr))
1766 newval = (uval & FUTEX_OWNER_DIED) | newtid;
1768 if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))
1776 * We fixed up user space. Now we need to fix the pi_state
1779 if (pi_state->owner != NULL) {
1780 raw_spin_lock_irq(&pi_state->owner->pi_lock);
1781 WARN_ON(list_empty(&pi_state->list));
1782 list_del_init(&pi_state->list);
1783 raw_spin_unlock_irq(&pi_state->owner->pi_lock);
1786 pi_state->owner = newowner;
1788 raw_spin_lock_irq(&newowner->pi_lock);
1789 WARN_ON(!list_empty(&pi_state->list));
1790 list_add(&pi_state->list, &newowner->pi_state_list);
1791 raw_spin_unlock_irq(&newowner->pi_lock);
1795 * To handle the page fault we need to drop the hash bucket
1796 * lock here. That gives the other task (either the highest priority
1797 * waiter itself or the task which stole the rtmutex) the
1798 * chance to try the fixup of the pi_state. So once we are
1799 * back from handling the fault we need to check the pi_state
1800 * after reacquiring the hash bucket lock and before trying to
1801 * do another fixup. When the fixup has been done already we
1805 spin_unlock(q->lock_ptr);
1807 ret = fault_in_user_writeable(uaddr);
1809 spin_lock(q->lock_ptr);
1812 * Check if someone else fixed it for us:
1814 if (pi_state->owner != oldowner)
1823 static long futex_wait_restart(struct restart_block *restart);
1826 * fixup_owner() - Post lock pi_state and corner case management
1827 * @uaddr: user address of the futex
1828 * @q: futex_q (contains pi_state and access to the rt_mutex)
1829 * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0)
1831 * After attempting to lock an rt_mutex, this function is called to cleanup
1832 * the pi_state owner as well as handle race conditions that may allow us to
1833 * acquire the lock. Must be called with the hb lock held.
1836 * 1 - success, lock taken
1837 * 0 - success, lock not taken
1838 * <0 - on error (-EFAULT)
1840 static int fixup_owner(u32 __user *uaddr, struct futex_q *q, int locked)
1842 struct task_struct *owner;
1847 * Got the lock. We might not be the anticipated owner if we
1848 * did a lock-steal - fix up the PI-state in that case:
1850 if (q->pi_state->owner != current)
1851 ret = fixup_pi_state_owner(uaddr, q, current);
1856 * Catch the rare case, where the lock was released when we were on the
1857 * way back before we locked the hash bucket.
1859 if (q->pi_state->owner == current) {
1861 * Try to get the rt_mutex now. This might fail as some other
1862 * task acquired the rt_mutex after we removed ourself from the
1863 * rt_mutex waiters list.
1865 if (rt_mutex_trylock(&q->pi_state->pi_mutex)) {
1871 * pi_state is incorrect, some other task did a lock steal and
1872 * we returned due to timeout or signal without taking the
1873 * rt_mutex. Too late.
1875 raw_spin_lock(&q->pi_state->pi_mutex.wait_lock);
1876 owner = rt_mutex_owner(&q->pi_state->pi_mutex);
1878 owner = rt_mutex_next_owner(&q->pi_state->pi_mutex);
1879 raw_spin_unlock(&q->pi_state->pi_mutex.wait_lock);
1880 ret = fixup_pi_state_owner(uaddr, q, owner);
1885 * Paranoia check. If we did not take the lock, then we should not be
1886 * the owner of the rt_mutex.
1888 if (rt_mutex_owner(&q->pi_state->pi_mutex) == current)
1889 printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p "
1890 "pi-state %p\n", ret,
1891 q->pi_state->pi_mutex.owner,
1892 q->pi_state->owner);
1895 return ret ? ret : locked;
1899 * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal
1900 * @hb: the futex hash bucket, must be locked by the caller
1901 * @q: the futex_q to queue up on
1902 * @timeout: the prepared hrtimer_sleeper, or null for no timeout
1904 static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q,
1905 struct hrtimer_sleeper *timeout)
1908 * The task state is guaranteed to be set before another task can
1909 * wake it. set_current_state() is implemented using set_mb() and
1910 * queue_me() calls spin_unlock() upon completion, both serializing
1911 * access to the hash list and forcing another memory barrier.
1913 set_current_state(TASK_INTERRUPTIBLE);
1918 hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS);
1919 if (!hrtimer_active(&timeout->timer))
1920 timeout->task = NULL;
1924 * If we have been removed from the hash list, then another task
1925 * has tried to wake us, and we can skip the call to schedule().
1927 if (likely(!plist_node_empty(&q->list))) {
1929 * If the timer has already expired, current will already be
1930 * flagged for rescheduling. Only call schedule if there
1931 * is no timeout, or if it has yet to expire.
1933 if (!timeout || timeout->task)
1936 __set_current_state(TASK_RUNNING);
1940 * futex_wait_setup() - Prepare to wait on a futex
1941 * @uaddr: the futex userspace address
1942 * @val: the expected value
1943 * @flags: futex flags (FLAGS_SHARED, etc.)
1944 * @q: the associated futex_q
1945 * @hb: storage for hash_bucket pointer to be returned to caller
1947 * Setup the futex_q and locate the hash_bucket. Get the futex value and
1948 * compare it with the expected value. Handle atomic faults internally.
1949 * Return with the hb lock held and a q.key reference on success, and unlocked
1950 * with no q.key reference on failure.
1953 * 0 - uaddr contains val and hb has been locked
1954 * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlocked
1956 static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags,
1957 struct futex_q *q, struct futex_hash_bucket **hb)
1963 * Access the page AFTER the hash-bucket is locked.
1964 * Order is important:
1966 * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val);
1967 * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); }
1969 * The basic logical guarantee of a futex is that it blocks ONLY
1970 * if cond(var) is known to be true at the time of blocking, for
1971 * any cond. If we locked the hash-bucket after testing *uaddr, that
1972 * would open a race condition where we could block indefinitely with
1973 * cond(var) false, which would violate the guarantee.
1975 * On the other hand, we insert q and release the hash-bucket only
1976 * after testing *uaddr. This guarantees that futex_wait() will NOT
1977 * absorb a wakeup if *uaddr does not match the desired values
1978 * while the syscall executes.
1981 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q->key, VERIFY_READ);
1982 if (unlikely(ret != 0))
1986 *hb = queue_lock(q);
1988 ret = get_futex_value_locked(&uval, uaddr);
1991 queue_unlock(q, *hb);
1993 ret = get_user(uval, uaddr);
1997 if (!(flags & FLAGS_SHARED))
2000 put_futex_key(&q->key);
2005 queue_unlock(q, *hb);
2011 put_futex_key(&q->key);
2015 static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val,
2016 ktime_t *abs_time, u32 bitset)
2018 struct hrtimer_sleeper timeout, *to = NULL;
2019 struct restart_block *restart;
2020 struct futex_hash_bucket *hb;
2021 struct futex_q q = futex_q_init;
2031 hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ?
2032 CLOCK_REALTIME : CLOCK_MONOTONIC,
2034 hrtimer_init_sleeper(to, current);
2035 hrtimer_set_expires_range_ns(&to->timer, *abs_time,
2036 current->timer_slack_ns);
2041 * Prepare to wait on uaddr. On success, holds hb lock and increments
2044 ret = futex_wait_setup(uaddr, val, flags, &q, &hb);
2048 /* queue_me and wait for wakeup, timeout, or a signal. */
2049 futex_wait_queue_me(hb, &q, to);
2051 /* If we were woken (and unqueued), we succeeded, whatever. */
2053 /* unqueue_me() drops q.key ref */
2054 if (!unqueue_me(&q))
2057 if (to && !to->task)
2061 * We expect signal_pending(current), but we might be the
2062 * victim of a spurious wakeup as well.
2064 if (!signal_pending(current))
2071 restart = ¤t_thread_info()->restart_block;
2072 restart->fn = futex_wait_restart;
2073 restart->futex.uaddr = uaddr;
2074 restart->futex.val = val;
2075 restart->futex.time = abs_time->tv64;
2076 restart->futex.bitset = bitset;
2077 restart->futex.flags = flags | FLAGS_HAS_TIMEOUT;
2079 ret = -ERESTART_RESTARTBLOCK;
2083 hrtimer_cancel(&to->timer);
2084 destroy_hrtimer_on_stack(&to->timer);
2090 static long futex_wait_restart(struct restart_block *restart)
2092 u32 __user *uaddr = restart->futex.uaddr;
2093 ktime_t t, *tp = NULL;
2095 if (restart->futex.flags & FLAGS_HAS_TIMEOUT) {
2096 t.tv64 = restart->futex.time;
2099 restart->fn = do_no_restart_syscall;
2101 return (long)futex_wait(uaddr, restart->futex.flags,
2102 restart->futex.val, tp, restart->futex.bitset);
2107 * Userspace tried a 0 -> TID atomic transition of the futex value
2108 * and failed. The kernel side here does the whole locking operation:
2109 * if there are waiters then it will block, it does PI, etc. (Due to
2110 * races the kernel might see a 0 value of the futex too.)
2112 static int futex_lock_pi(u32 __user *uaddr, unsigned int flags, int detect,
2113 ktime_t *time, int trylock)
2115 struct hrtimer_sleeper timeout, *to = NULL;
2116 struct futex_hash_bucket *hb;
2117 struct futex_q q = futex_q_init;
2120 if (refill_pi_state_cache())
2125 hrtimer_init_on_stack(&to->timer, CLOCK_REALTIME,
2127 hrtimer_init_sleeper(to, current);
2128 hrtimer_set_expires(&to->timer, *time);
2132 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q.key, VERIFY_WRITE);
2133 if (unlikely(ret != 0))
2137 hb = queue_lock(&q);
2139 ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, 0);
2140 if (unlikely(ret)) {
2143 /* We got the lock. */
2145 goto out_unlock_put_key;
2150 * Task is exiting and we just wait for the
2153 queue_unlock(&q, hb);
2154 put_futex_key(&q.key);
2158 goto out_unlock_put_key;
2163 * Only actually queue now that the atomic ops are done:
2167 WARN_ON(!q.pi_state);
2169 * Block on the PI mutex:
2172 ret = rt_mutex_timed_lock(&q.pi_state->pi_mutex, to, 1);
2174 ret = rt_mutex_trylock(&q.pi_state->pi_mutex);
2175 /* Fixup the trylock return value: */
2176 ret = ret ? 0 : -EWOULDBLOCK;
2179 spin_lock(q.lock_ptr);
2181 * Fixup the pi_state owner and possibly acquire the lock if we
2184 res = fixup_owner(uaddr, &q, !ret);
2186 * If fixup_owner() returned an error, proprogate that. If it acquired
2187 * the lock, clear our -ETIMEDOUT or -EINTR.
2190 ret = (res < 0) ? res : 0;
2193 * If fixup_owner() faulted and was unable to handle the fault, unlock
2194 * it and return the fault to userspace.
2196 if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current))
2197 rt_mutex_unlock(&q.pi_state->pi_mutex);
2199 /* Unqueue and drop the lock */
2205 queue_unlock(&q, hb);
2208 put_futex_key(&q.key);
2211 destroy_hrtimer_on_stack(&to->timer);
2212 return ret != -EINTR ? ret : -ERESTARTNOINTR;
2215 queue_unlock(&q, hb);
2217 ret = fault_in_user_writeable(uaddr);
2221 if (!(flags & FLAGS_SHARED))
2224 put_futex_key(&q.key);
2229 * Userspace attempted a TID -> 0 atomic transition, and failed.
2230 * This is the in-kernel slowpath: we look up the PI state (if any),
2231 * and do the rt-mutex unlock.
2233 static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags)
2235 struct futex_hash_bucket *hb;
2236 struct futex_q *this, *next;
2237 struct plist_head *head;
2238 union futex_key key = FUTEX_KEY_INIT;
2239 u32 uval, vpid = task_pid_vnr(current);
2243 if (get_user(uval, uaddr))
2246 * We release only a lock we actually own:
2248 if ((uval & FUTEX_TID_MASK) != vpid)
2251 ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, VERIFY_WRITE);
2252 if (unlikely(ret != 0))
2255 hb = hash_futex(&key);
2256 spin_lock(&hb->lock);
2259 * To avoid races, try to do the TID -> 0 atomic transition
2260 * again. If it succeeds then we can return without waking
2261 * anyone else up. We only try this if neither the waiters nor
2262 * the owner died bit are set.
2264 if (!(uval & ~FUTEX_TID_MASK) &&
2265 cmpxchg_futex_value_locked(&uval, uaddr, vpid, 0))
2268 * Rare case: we managed to release the lock atomically,
2269 * no need to wake anyone else up:
2271 if (unlikely(uval == vpid))
2275 * Ok, other tasks may need to be woken up - check waiters
2276 * and do the wakeup if necessary:
2280 plist_for_each_entry_safe(this, next, head, list) {
2281 if (!match_futex (&this->key, &key))
2283 ret = wake_futex_pi(uaddr, uval, this);
2285 * The atomic access to the futex value
2286 * generated a pagefault, so retry the
2287 * user-access and the wakeup:
2294 * No waiters - kernel unlocks the futex:
2296 ret = unlock_futex_pi(uaddr, uval);
2301 spin_unlock(&hb->lock);
2302 put_futex_key(&key);
2308 spin_unlock(&hb->lock);
2309 put_futex_key(&key);
2311 ret = fault_in_user_writeable(uaddr);
2319 * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex
2320 * @hb: the hash_bucket futex_q was original enqueued on
2321 * @q: the futex_q woken while waiting to be requeued
2322 * @key2: the futex_key of the requeue target futex
2323 * @timeout: the timeout associated with the wait (NULL if none)
2325 * Detect if the task was woken on the initial futex as opposed to the requeue
2326 * target futex. If so, determine if it was a timeout or a signal that caused
2327 * the wakeup and return the appropriate error code to the caller. Must be
2328 * called with the hb lock held.
2331 * 0 - no early wakeup detected
2332 * <0 - -ETIMEDOUT or -ERESTARTNOINTR
2335 int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb,
2336 struct futex_q *q, union futex_key *key2,
2337 struct hrtimer_sleeper *timeout)
2342 * With the hb lock held, we avoid races while we process the wakeup.
2343 * We only need to hold hb (and not hb2) to ensure atomicity as the
2344 * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb.
2345 * It can't be requeued from uaddr2 to something else since we don't
2346 * support a PI aware source futex for requeue.
2348 if (!match_futex(&q->key, key2)) {
2349 WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr));
2351 * We were woken prior to requeue by a timeout or a signal.
2352 * Unqueue the futex_q and determine which it was.
2354 plist_del(&q->list, &hb->chain);
2356 /* Handle spurious wakeups gracefully */
2358 if (timeout && !timeout->task)
2360 else if (signal_pending(current))
2361 ret = -ERESTARTNOINTR;
2367 * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2
2368 * @uaddr: the futex we initially wait on (non-pi)
2369 * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be
2370 * the same type, no requeueing from private to shared, etc.
2371 * @val: the expected value of uaddr
2372 * @abs_time: absolute timeout
2373 * @bitset: 32 bit wakeup bitset set by userspace, defaults to all
2374 * @clockrt: whether to use CLOCK_REALTIME (1) or CLOCK_MONOTONIC (0)
2375 * @uaddr2: the pi futex we will take prior to returning to user-space
2377 * The caller will wait on uaddr and will be requeued by futex_requeue() to
2378 * uaddr2 which must be PI aware and unique from uaddr. Normal wakeup will wake
2379 * on uaddr2 and complete the acquisition of the rt_mutex prior to returning to
2380 * userspace. This ensures the rt_mutex maintains an owner when it has waiters;
2381 * without one, the pi logic would not know which task to boost/deboost, if
2382 * there was a need to.
2384 * We call schedule in futex_wait_queue_me() when we enqueue and return there
2385 * via the following:
2386 * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue()
2387 * 2) wakeup on uaddr2 after a requeue
2391 * If 3, cleanup and return -ERESTARTNOINTR.
2393 * If 2, we may then block on trying to take the rt_mutex and return via:
2394 * 5) successful lock
2397 * 8) other lock acquisition failure
2399 * If 6, return -EWOULDBLOCK (restarting the syscall would do the same).
2401 * If 4 or 7, we cleanup and return with -ETIMEDOUT.
2407 static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
2408 u32 val, ktime_t *abs_time, u32 bitset,
2411 struct hrtimer_sleeper timeout, *to = NULL;
2412 struct rt_mutex_waiter rt_waiter;
2413 struct rt_mutex *pi_mutex = NULL;
2414 struct futex_hash_bucket *hb;
2415 union futex_key key2 = FUTEX_KEY_INIT;
2416 struct futex_q q = futex_q_init;
2419 if (uaddr == uaddr2)
2427 hrtimer_init_on_stack(&to->timer, (flags & FLAGS_CLOCKRT) ?
2428 CLOCK_REALTIME : CLOCK_MONOTONIC,
2430 hrtimer_init_sleeper(to, current);
2431 hrtimer_set_expires_range_ns(&to->timer, *abs_time,
2432 current->timer_slack_ns);
2436 * The waiter is allocated on our stack, manipulated by the requeue
2437 * code while we sleep on uaddr.
2439 debug_rt_mutex_init_waiter(&rt_waiter);
2440 rt_waiter.task = NULL;
2442 ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, VERIFY_WRITE);
2443 if (unlikely(ret != 0))
2447 q.rt_waiter = &rt_waiter;
2448 q.requeue_pi_key = &key2;
2451 * Prepare to wait on uaddr. On success, increments q.key (key1) ref
2454 ret = futex_wait_setup(uaddr, val, flags, &q, &hb);
2459 * The check above which compares uaddrs is not sufficient for
2460 * shared futexes. We need to compare the keys:
2462 if (match_futex(&q.key, &key2)) {
2463 queue_unlock(&q, hb);
2468 /* Queue the futex_q, drop the hb lock, wait for wakeup. */
2469 futex_wait_queue_me(hb, &q, to);
2471 spin_lock(&hb->lock);
2472 ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to);
2473 spin_unlock(&hb->lock);
2478 * In order for us to be here, we know our q.key == key2, and since
2479 * we took the hb->lock above, we also know that futex_requeue() has
2480 * completed and we no longer have to concern ourselves with a wakeup
2481 * race with the atomic proxy lock acquisition by the requeue code. The
2482 * futex_requeue dropped our key1 reference and incremented our key2
2486 /* Check if the requeue code acquired the second futex for us. */
2489 * Got the lock. We might not be the anticipated owner if we
2490 * did a lock-steal - fix up the PI-state in that case.
2492 if (q.pi_state && (q.pi_state->owner != current)) {
2493 spin_lock(q.lock_ptr);
2494 ret = fixup_pi_state_owner(uaddr2, &q, current);
2496 * Drop the reference to the pi state which
2497 * the requeue_pi() code acquired for us.
2499 free_pi_state(q.pi_state);
2500 spin_unlock(q.lock_ptr);
2504 * We have been woken up by futex_unlock_pi(), a timeout, or a
2505 * signal. futex_unlock_pi() will not destroy the lock_ptr nor
2508 WARN_ON(!q.pi_state);
2509 pi_mutex = &q.pi_state->pi_mutex;
2510 ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter, 1);
2511 debug_rt_mutex_free_waiter(&rt_waiter);
2513 spin_lock(q.lock_ptr);
2515 * Fixup the pi_state owner and possibly acquire the lock if we
2518 res = fixup_owner(uaddr2, &q, !ret);
2520 * If fixup_owner() returned an error, proprogate that. If it
2521 * acquired the lock, clear -ETIMEDOUT or -EINTR.
2524 ret = (res < 0) ? res : 0;
2526 /* Unqueue and drop the lock. */
2531 * If fixup_pi_state_owner() faulted and was unable to handle the
2532 * fault, unlock the rt_mutex and return the fault to userspace.
2534 if (ret == -EFAULT) {
2535 if (pi_mutex && rt_mutex_owner(pi_mutex) == current)
2536 rt_mutex_unlock(pi_mutex);
2537 } else if (ret == -EINTR) {
2539 * We've already been requeued, but cannot restart by calling
2540 * futex_lock_pi() directly. We could restart this syscall, but
2541 * it would detect that the user space "val" changed and return
2542 * -EWOULDBLOCK. Save the overhead of the restart and return
2543 * -EWOULDBLOCK directly.
2549 put_futex_key(&q.key);
2551 put_futex_key(&key2);
2555 hrtimer_cancel(&to->timer);
2556 destroy_hrtimer_on_stack(&to->timer);
2562 * Support for robust futexes: the kernel cleans up held futexes at
2565 * Implementation: user-space maintains a per-thread list of locks it
2566 * is holding. Upon do_exit(), the kernel carefully walks this list,
2567 * and marks all locks that are owned by this thread with the
2568 * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is
2569 * always manipulated with the lock held, so the list is private and
2570 * per-thread. Userspace also maintains a per-thread 'list_op_pending'
2571 * field, to allow the kernel to clean up if the thread dies after
2572 * acquiring the lock, but just before it could have added itself to
2573 * the list. There can only be one such pending lock.
2577 * sys_set_robust_list() - Set the robust-futex list head of a task
2578 * @head: pointer to the list-head
2579 * @len: length of the list-head, as userspace expects
2581 SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head,
2584 if (!futex_cmpxchg_enabled)
2587 * The kernel knows only one size for now:
2589 if (unlikely(len != sizeof(*head)))
2592 current->robust_list = head;
2598 * sys_get_robust_list() - Get the robust-futex list head of a task
2599 * @pid: pid of the process [zero for current task]
2600 * @head_ptr: pointer to a list-head pointer, the kernel fills it in
2601 * @len_ptr: pointer to a length field, the kernel fills in the header size
2603 SYSCALL_DEFINE3(get_robust_list, int, pid,
2604 struct robust_list_head __user * __user *, head_ptr,
2605 size_t __user *, len_ptr)
2607 struct robust_list_head __user *head;
2609 struct task_struct *p;
2611 if (!futex_cmpxchg_enabled)
2620 p = find_task_by_vpid(pid);
2626 if (!ptrace_may_access(p, PTRACE_MODE_READ))
2629 head = p->robust_list;
2632 if (put_user(sizeof(*head), len_ptr))
2634 return put_user(head, head_ptr);
2643 * Process a futex-list entry, check whether it's owned by the
2644 * dying task, and do notification if so:
2646 int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi)
2648 u32 uval, uninitialized_var(nval), mval;
2651 if (get_user(uval, uaddr))
2654 if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) {
2656 * Ok, this dying thread is truly holding a futex
2657 * of interest. Set the OWNER_DIED bit atomically
2658 * via cmpxchg, and if the value had FUTEX_WAITERS
2659 * set, wake up a waiter (if any). (We have to do a
2660 * futex_wake() even if OWNER_DIED is already set -
2661 * to handle the rare but possible case of recursive
2662 * thread-death.) The rest of the cleanup is done in
2665 mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED;
2667 * We are not holding a lock here, but we want to have
2668 * the pagefault_disable/enable() protection because
2669 * we want to handle the fault gracefully. If the
2670 * access fails we try to fault in the futex with R/W
2671 * verification via get_user_pages. get_user() above
2672 * does not guarantee R/W access. If that fails we
2673 * give up and leave the futex locked.
2675 if (cmpxchg_futex_value_locked(&nval, uaddr, uval, mval)) {
2676 if (fault_in_user_writeable(uaddr))
2684 * Wake robust non-PI futexes here. The wakeup of
2685 * PI futexes happens in exit_pi_state():
2687 if (!pi && (uval & FUTEX_WAITERS))
2688 futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY);
2694 * Fetch a robust-list pointer. Bit 0 signals PI futexes:
2696 static inline int fetch_robust_entry(struct robust_list __user **entry,
2697 struct robust_list __user * __user *head,
2700 unsigned long uentry;
2702 if (get_user(uentry, (unsigned long __user *)head))
2705 *entry = (void __user *)(uentry & ~1UL);
2712 * Walk curr->robust_list (very carefully, it's a userspace list!)
2713 * and mark any locks found there dead, and notify any waiters.
2715 * We silently return on any sign of list-walking problem.
2717 void exit_robust_list(struct task_struct *curr)
2719 struct robust_list_head __user *head = curr->robust_list;
2720 struct robust_list __user *entry, *next_entry, *pending;
2721 unsigned int limit = ROBUST_LIST_LIMIT, pi, pip;
2722 unsigned int uninitialized_var(next_pi);
2723 unsigned long futex_offset;
2726 if (!futex_cmpxchg_enabled)
2730 * Fetch the list head (which was registered earlier, via
2731 * sys_set_robust_list()):
2733 if (fetch_robust_entry(&entry, &head->list.next, &pi))
2736 * Fetch the relative futex offset:
2738 if (get_user(futex_offset, &head->futex_offset))
2741 * Fetch any possibly pending lock-add first, and handle it
2744 if (fetch_robust_entry(&pending, &head->list_op_pending, &pip))
2747 next_entry = NULL; /* avoid warning with gcc */
2748 while (entry != &head->list) {
2750 * Fetch the next entry in the list before calling
2751 * handle_futex_death:
2753 rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi);
2755 * A pending lock might already be on the list, so
2756 * don't process it twice:
2758 if (entry != pending)
2759 if (handle_futex_death((void __user *)entry + futex_offset,
2767 * Avoid excessively long or circular lists:
2776 handle_futex_death((void __user *)pending + futex_offset,
2780 long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout,
2781 u32 __user *uaddr2, u32 val2, u32 val3)
2783 int ret = -ENOSYS, cmd = op & FUTEX_CMD_MASK;
2784 unsigned int flags = 0;
2786 if (!(op & FUTEX_PRIVATE_FLAG))
2787 flags |= FLAGS_SHARED;
2789 if (op & FUTEX_CLOCK_REALTIME) {
2790 flags |= FLAGS_CLOCKRT;
2791 if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI)
2797 case FUTEX_UNLOCK_PI:
2798 case FUTEX_TRYLOCK_PI:
2799 case FUTEX_WAIT_REQUEUE_PI:
2800 case FUTEX_CMP_REQUEUE_PI:
2801 if (!futex_cmpxchg_enabled)
2807 val3 = FUTEX_BITSET_MATCH_ANY;
2808 case FUTEX_WAIT_BITSET:
2809 ret = futex_wait(uaddr, flags, val, timeout, val3);
2812 val3 = FUTEX_BITSET_MATCH_ANY;
2813 case FUTEX_WAKE_BITSET:
2814 ret = futex_wake(uaddr, flags, val, val3);
2817 ret = futex_requeue(uaddr, flags, uaddr2, val, val2, NULL, 0);
2819 case FUTEX_CMP_REQUEUE:
2820 ret = futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 0);
2823 ret = futex_wake_op(uaddr, flags, uaddr2, val, val2, val3);
2826 ret = futex_lock_pi(uaddr, flags, val, timeout, 0);
2828 case FUTEX_UNLOCK_PI:
2829 ret = futex_unlock_pi(uaddr, flags);
2831 case FUTEX_TRYLOCK_PI:
2832 ret = futex_lock_pi(uaddr, flags, 0, timeout, 1);
2834 case FUTEX_WAIT_REQUEUE_PI:
2835 val3 = FUTEX_BITSET_MATCH_ANY;
2836 ret = futex_wait_requeue_pi(uaddr, flags, val, timeout, val3,
2839 case FUTEX_CMP_REQUEUE_PI:
2840 ret = futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 1);
2849 SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val,
2850 struct timespec __user *, utime, u32 __user *, uaddr2,
2854 ktime_t t, *tp = NULL;
2856 int cmd = op & FUTEX_CMD_MASK;
2858 if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI ||
2859 cmd == FUTEX_WAIT_BITSET ||
2860 cmd == FUTEX_WAIT_REQUEUE_PI)) {
2861 if (copy_from_user(&ts, utime, sizeof(ts)) != 0)
2863 if (!timespec_valid(&ts))
2866 t = timespec_to_ktime(ts);
2867 if (cmd == FUTEX_WAIT)
2868 t = ktime_add_safe(ktime_get(), t);
2872 * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*.
2873 * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP.
2875 if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE ||
2876 cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP)
2877 val2 = (u32) (unsigned long) utime;
2879 return do_futex(uaddr, op, val, tp, uaddr2, val2, val3);
2882 static int __init futex_init(void)
2888 * This will fail and we want it. Some arch implementations do
2889 * runtime detection of the futex_atomic_cmpxchg_inatomic()
2890 * functionality. We want to know that before we call in any
2891 * of the complex code paths. Also we want to prevent
2892 * registration of robust lists in that case. NULL is
2893 * guaranteed to fault and we get -EFAULT on functional
2894 * implementation, the non-functional ones will return
2897 if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT)
2898 futex_cmpxchg_enabled = 1;
2900 for (i = 0; i < ARRAY_SIZE(futex_queues); i++) {
2901 plist_head_init(&futex_queues[i].chain);
2902 spin_lock_init(&futex_queues[i].lock);
2907 core_initcall(futex_init);