History

futex: Rework futex_lock_pi() to use rt_mutex_*_proxy_lock()

By changing futex_lock_pi() to use rt_mutex_*_proxy_lock() all wait_list modifications are done under both hb->lock and wait_lock. This closes the obvious interleave pattern between futex_lock_pi() and futex_unlock_pi(), but not entirely so. See below: Before: futex_lock_pi() futex_unlock_pi() unlock hb->lock lock hb->lock unlock hb->lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock schedule() lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock <idem> -EAGAIN lock hb->lock After: futex_lock_pi() futex_unlock_pi() lock hb->lock lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock unlock hb->lock schedule() lock hb->lock unlock hb->lock lock hb->lock lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN unlock hb->lock It does however solve the earlier starvation/live-lock scenario which got introduced with the -EAGAIN since unlike the before scenario; where the -EAGAIN happens while futex_unlock_pi() doesn't hold any locks; in the after scenario it happens while futex_unlock_pi() actually holds a lock, and then it is serialized on that lock.

Peter Zijlstra committed 9 years ago

Commit cfafcd11 parent 38d589f3 files · +62 −42Browse files

Unified · Split

3 files
  1. kernel/futex.c +55 −22
  2. kernel/locking/rtmutex.c +7 −19
  3. kernel/locking/rtmutex_common.h +0 −1
kernel/futex.c+55−22
@@ -2099,20 +2099,7 @@ queue_unlock(struct futex_hash_bucket *hb) · 2098 unmodified lines
20992099 hb_waiters_dec(hb);
21002100 }
21012101
2102−/**
2103− * queue_me() - Enqueue the futex_q on the futex_hash_bucket
2104− * @q: The futex_q to enqueue
2105− * @hb: The destination hash bucket
2106− *
2107− * The hb->lock must be held by the caller, and is released here. A call to
2108− * queue_me() is typically paired with exactly one call to unqueue_me(). The
2109− * exceptions involve the PI related operations, which may use unqueue_me_pi()
2110− * or nothing if the unqueue is done as part of the wake process and the unqueue
2111− * state is implicit in the state of woken task (see futex_wait_requeue_pi() for
2112− * an example).
2113− */
2114−static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb)
2115− __releases(&hb->lock)
2102+static inline void __queue_me(struct futex_q *q, struct futex_hash_bucket *hb)
21162103 {
21172104 int prio;
21182105
@@ -2129,6 +2116,24 @@ static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb) · 10 unmodified lines
21292116 plist_node_init(&q->list, prio);
21302117 plist_add(&q->list, &hb->chain);
21312118 q->task = current;
2119+}
2120+
2121+/**
2122+ * queue_me() - Enqueue the futex_q on the futex_hash_bucket
2123+ * @q: The futex_q to enqueue
2124+ * @hb: The destination hash bucket
2125+ *
2126+ * The hb->lock must be held by the caller, and is released here. A call to
2127+ * queue_me() is typically paired with exactly one call to unqueue_me(). The
2128+ * exceptions involve the PI related operations, which may use unqueue_me_pi()
2129+ * or nothing if the unqueue is done as part of the wake process and the unqueue
2130+ * state is implicit in the state of woken task (see futex_wait_requeue_pi() for
2131+ * an example).
2132+ */
2133+static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb)
2134+ __releases(&hb->lock)
2135+{
2136+ __queue_me(q, hb);
21322137 spin_unlock(&hb->lock);
21332138 }
21342139
@@ -2587,6 +2592,7 @@ static int futex_lock_pi(u32 __user *uaddr, unsigned int flags, · 452 unmodified lines
25872592 {
25882593 struct hrtimer_sleeper timeout, *to = NULL;
25892594 struct futex_pi_state *pi_state = NULL;
2595+ struct rt_mutex_waiter rt_waiter;
25902596 struct futex_hash_bucket *hb;
25912597 struct futex_q q = futex_q_init;
25922598 int res, ret;
@@ -2639,24 +2645,51 @@ retry_private: · 46 unmodified lines
26392645 }
26402646 }
26412647
2648+ WARN_ON(!q.pi_state);
2649+
26422650 /*
26432651 * Only actually queue now that the atomic ops are done:
26442652 */
2645− queue_me(&q, hb);
2653+ __queue_me(&q, hb);
26462654
2647− WARN_ON(!q.pi_state);
2648− /*
2649− * Block on the PI mutex:
2650− */
2651− if (!trylock) {
2652− ret = rt_mutex_timed_futex_lock(&q.pi_state->pi_mutex, to);
2653− } else {
2655+ if (trylock) {
26542656 ret = rt_mutex_futex_trylock(&q.pi_state->pi_mutex);
26552657 /* Fixup the trylock return value: */
26562658 ret = ret ? 0 : -EWOULDBLOCK;
2659+ goto no_block;
2660+ }
2661+
2662+ /*
2663+ * We must add ourselves to the rt_mutex waitlist while holding hb->lock
2664+ * such that the hb and rt_mutex wait lists match.
2665+ */
2666+ rt_mutex_init_waiter(&rt_waiter);
2667+ ret = rt_mutex_start_proxy_lock(&q.pi_state->pi_mutex, &rt_waiter, current);
2668+ if (ret) {
2669+ if (ret == 1)
2670+ ret = 0;
2671+
2672+ goto no_block;
26572673 }
26582674
2675+ spin_unlock(q.lock_ptr);
2676+
2677+ if (unlikely(to))
2678+ hrtimer_start_expires(&to->timer, HRTIMER_MODE_ABS);
2679+
2680+ ret = rt_mutex_wait_proxy_lock(&q.pi_state->pi_mutex, to, &rt_waiter);
2681+
26592682 spin_lock(q.lock_ptr);
2683+ /*
2684+ * If we failed to acquire the lock (signal/timeout), we must
2685+ * first acquire the hb->lock before removing the lock from the
2686+ * rt_mutex waitqueue, such that we can keep the hb and rt_mutex
2687+ * wait lists consistent.
2688+ */
2689+ if (ret && !rt_mutex_cleanup_proxy_lock(&q.pi_state->pi_mutex, &rt_waiter))
2690+ ret = 0;
2691+
2692+no_block:
26602693 /*
26612694 * Fixup the pi_state owner and possibly acquire the lock if we
26622695 * haven't already.
kernel/locking/rtmutex.c+7−19
@@ -1492,19 +1492,6 @@ int __sched rt_mutex_lock_interruptible(struct rt_mutex *lock) · 1491 unmodified lines
14921492 }
14931493 EXPORT_SYMBOL_GPL(rt_mutex_lock_interruptible);
14941494
1495−/*
1496− * Futex variant with full deadlock detection.
1497− * Futex variants must not use the fast-path, see __rt_mutex_futex_unlock().
1498− */
1499−int __sched rt_mutex_timed_futex_lock(struct rt_mutex *lock,
1500− struct hrtimer_sleeper *timeout)
1501−{
1502− might_sleep();
1503−
1504− return rt_mutex_slowlock(lock, TASK_INTERRUPTIBLE,
1505− timeout, RT_MUTEX_FULL_CHAINWALK);
1506−}
1507−
15081495 /*
15091496 * Futex variant, must not use fastpath.
15101497 */
@@ -1782,12 +1769,6 @@ int rt_mutex_wait_proxy_lock(struct rt_mutex *lock, · 271 unmodified lines
17821769 /* sleep on the mutex */
17831770 ret = __rt_mutex_slowlock(lock, TASK_INTERRUPTIBLE, to, waiter);
17841771
1785− /*
1786− * try_to_take_rt_mutex() sets the waiter bit unconditionally. We might
1787− * have to fix that up.
1788− */
1789− fixup_rt_mutex_waiters(lock);
1790−
17911772 raw_spin_unlock_irq(&lock->wait_lock);
17921773
17931774 return ret;
@@ -1827,6 +1808,13 @@ bool rt_mutex_cleanup_proxy_lock(struct rt_mutex *lock, · 33 unmodified lines
18271808 fixup_rt_mutex_waiters(lock);
18281809 cleanup = true;
18291810 }
1811+
1812+ /*
1813+ * try_to_take_rt_mutex() sets the waiter bit unconditionally. We might
1814+ * have to fix that up.
1815+ */
1816+ fixup_rt_mutex_waiters(lock);
1817+
18301818 raw_spin_unlock_irq(&lock->wait_lock);
18311819
18321820 return cleanup;
kernel/locking/rtmutex_common.h+0−1
@@ -113,7 +113,6 @@ extern int rt_mutex_wait_proxy_lock(struct rt_mutex *lock, · 112 unmodified lines
113113 extern bool rt_mutex_cleanup_proxy_lock(struct rt_mutex *lock,
114114 struct rt_mutex_waiter *waiter);
115115
116−extern int rt_mutex_timed_futex_lock(struct rt_mutex *l, struct hrtimer_sleeper *to);
117116 extern int rt_mutex_futex_trylock(struct rt_mutex *l);
118117
119118 extern void rt_mutex_futex_unlock(struct rt_mutex *lock);

Keyboard shortcuts

General

Show keyboard shortcuts
?
Focus the current search
/

Issues and pull requests

Open filters
f

Issues, pull requests and Inbox

Select next item
j↓
Select previous item
k↑
Open selected item
Enter
Dismiss filters or clear selection
Esc

Repository code

Go to file
t