parking_lot/
fair_mutex.rs

1// Copyright 2016 Amanieu d'Antras
2//
3// Licensed under the Apache License, Version 2.0, <LICENSE-APACHE or
4// http://apache.org/licenses/LICENSE-2.0> or the MIT license <LICENSE-MIT or
5// http://opensource.org/licenses/MIT>, at your option. This file may not be
6// copied, modified, or distributed except according to those terms.
7
8use crate::raw_fair_mutex::RawFairMutex;
9
10/// A mutual exclusive primitive that is always fair, useful for protecting shared data
11///
12/// This mutex will block threads waiting for the lock to become available. The
13/// mutex can be statically initialized or created by the `new`
14/// constructor. Each mutex has a type parameter which represents the data that
15/// it is protecting. The data can only be accessed through the RAII guards
16/// returned from `lock` and `try_lock`, which guarantees that the data is only
17/// ever accessed when the mutex is locked.
18///
19/// The regular mutex provided by `parking_lot` uses eventual fairness
20/// (after some time it will default to the fair algorithm), but eventual
21/// fairness does not provide the same guarantees an always fair method would.
22/// Fair mutexes are generally slower, but sometimes needed.
23///
24/// In a fair mutex the waiters form a queue, and the lock is always granted to
25/// the next requester in the queue, in first-in first-out order. This ensures
26/// that one thread cannot starve others by quickly re-acquiring the lock after
27/// releasing it.
28///
29/// A fair mutex may not be interesting if threads have different priorities (this is known as
30/// priority inversion).
31///
32/// # Differences from the standard library `Mutex`
33///
34/// - No poisoning, the lock is released normally on panic.
35/// - Only requires 1 byte of space, whereas the standard library boxes the
36///   `FairMutex` due to platform limitations.
37/// - Can be statically constructed.
38/// - Does not require any drop glue when dropped.
39/// - Inline fast path for the uncontended case.
40/// - Efficient handling of micro-contention using adaptive spinning.
41/// - Allows raw locking & unlocking without a guard.
42///
43/// # Examples
44///
45/// ```
46/// use parking_lot::FairMutex;
47/// use std::sync::{Arc, mpsc::channel};
48/// use std::thread;
49///
50/// const N: usize = 10;
51///
52/// // Spawn a few threads to increment a shared variable (non-atomically), and
53/// // let the main thread know once all increments are done.
54/// //
55/// // Here we're using an Arc to share memory among threads, and the data inside
56/// // the Arc is protected with a mutex.
57/// let data = Arc::new(FairMutex::new(0));
58///
59/// let (tx, rx) = channel();
60/// for _ in 0..10 {
61///     let (data, tx) = (Arc::clone(&data), tx.clone());
62///     thread::spawn(move || {
63///         // The shared state can only be accessed once the lock is held.
64///         // Our non-atomic increment is safe because we're the only thread
65///         // which can access the shared state when the lock is held.
66///         let mut data = data.lock();
67///         *data += 1;
68///         if *data == N {
69///             tx.send(()).unwrap();
70///         }
71///         // the lock is unlocked here when `data` goes out of scope.
72///     });
73/// }
74///
75/// rx.recv().unwrap();
76/// ```
77pub type FairMutex<T> = lock_api::Mutex<RawFairMutex, T>;
78
79/// Creates a new fair mutex in an unlocked state ready for use.
80///
81/// This allows creating a fair mutex in a constant context on stable Rust.
82pub const fn const_fair_mutex<T>(val: T) -> FairMutex<T> {
83    FairMutex::const_new(<RawFairMutex as lock_api::RawMutex>::INIT, val)
84}
85
86/// An RAII implementation of a "scoped lock" of a mutex. When this structure is
87/// dropped (falls out of scope), the lock will be unlocked.
88///
89/// The data protected by the mutex can be accessed through this guard via its
90/// `Deref` and `DerefMut` implementations.
91pub type FairMutexGuard<'a, T> = lock_api::MutexGuard<'a, RawFairMutex, T>;
92
93/// An RAII mutex guard returned by `FairMutexGuard::map`, which can point to a
94/// subfield of the protected data.
95///
96/// The main difference between `MappedFairMutexGuard` and `FairMutexGuard` is that the
97/// former doesn't support temporarily unlocking and re-locking, since that
98/// could introduce soundness issues if the locked object is modified by another
99/// thread.
100pub type MappedFairMutexGuard<'a, T> = lock_api::MappedMutexGuard<'a, RawFairMutex, T>;
101
102#[cfg(test)]
103mod tests {
104    use crate::FairMutex;
105    use std::sync::atomic::{AtomicUsize, Ordering};
106    use std::sync::mpsc::channel;
107    use std::sync::Arc;
108    use std::thread;
109
110    #[cfg(feature = "serde")]
111    use bincode::{deserialize, serialize};
112
113    #[derive(Eq, PartialEq, Debug)]
114    struct NonCopy(i32);
115
116    #[test]
117    fn smoke() {
118        let m = FairMutex::new(());
119        drop(m.lock());
120        drop(m.lock());
121    }
122
123    #[test]
124    fn lots_and_lots() {
125        const J: u32 = 1000;
126        const K: u32 = 3;
127
128        let m = Arc::new(FairMutex::new(0));
129
130        fn inc(m: &FairMutex<u32>) {
131            for _ in 0..J {
132                *m.lock() += 1;
133            }
134        }
135
136        let (tx, rx) = channel();
137        for _ in 0..K {
138            let tx2 = tx.clone();
139            let m2 = m.clone();
140            thread::spawn(move || {
141                inc(&m2);
142                tx2.send(()).unwrap();
143            });
144            let tx2 = tx.clone();
145            let m2 = m.clone();
146            thread::spawn(move || {
147                inc(&m2);
148                tx2.send(()).unwrap();
149            });
150        }
151
152        drop(tx);
153        for _ in 0..2 * K {
154            rx.recv().unwrap();
155        }
156        assert_eq!(*m.lock(), J * K * 2);
157    }
158
159    #[test]
160    fn try_lock() {
161        let m = FairMutex::new(());
162        *m.try_lock().unwrap() = ();
163    }
164
165    #[test]
166    fn test_into_inner() {
167        let m = FairMutex::new(NonCopy(10));
168        assert_eq!(m.into_inner(), NonCopy(10));
169    }
170
171    #[test]
172    fn test_into_inner_drop() {
173        struct Foo(Arc<AtomicUsize>);
174        impl Drop for Foo {
175            fn drop(&mut self) {
176                self.0.fetch_add(1, Ordering::SeqCst);
177            }
178        }
179        let num_drops = Arc::new(AtomicUsize::new(0));
180        let m = FairMutex::new(Foo(num_drops.clone()));
181        assert_eq!(num_drops.load(Ordering::SeqCst), 0);
182        {
183            let _inner = m.into_inner();
184            assert_eq!(num_drops.load(Ordering::SeqCst), 0);
185        }
186        assert_eq!(num_drops.load(Ordering::SeqCst), 1);
187    }
188
189    #[test]
190    fn test_get_mut() {
191        let mut m = FairMutex::new(NonCopy(10));
192        *m.get_mut() = NonCopy(20);
193        assert_eq!(m.into_inner(), NonCopy(20));
194    }
195
196    #[test]
197    fn test_mutex_arc_nested() {
198        // Tests nested mutexes and access
199        // to underlying data.
200        let arc = Arc::new(FairMutex::new(1));
201        let arc2 = Arc::new(FairMutex::new(arc));
202        let (tx, rx) = channel();
203        let _t = thread::spawn(move || {
204            let lock = arc2.lock();
205            let lock2 = lock.lock();
206            assert_eq!(*lock2, 1);
207            tx.send(()).unwrap();
208        });
209        rx.recv().unwrap();
210    }
211
212    #[test]
213    fn test_mutex_arc_access_in_unwind() {
214        let arc = Arc::new(FairMutex::new(1));
215        let arc2 = arc.clone();
216        let _ = thread::spawn(move || {
217            struct Unwinder {
218                i: Arc<FairMutex<i32>>,
219            }
220            impl Drop for Unwinder {
221                fn drop(&mut self) {
222                    *self.i.lock() += 1;
223                }
224            }
225            let _u = Unwinder { i: arc2 };
226            panic!();
227        })
228        .join();
229        let lock = arc.lock();
230        assert_eq!(*lock, 2);
231    }
232
233    #[test]
234    fn test_mutex_unsized() {
235        let mutex: &FairMutex<[i32]> = &FairMutex::new([1, 2, 3]);
236        {
237            let b = &mut *mutex.lock();
238            b[0] = 4;
239            b[2] = 5;
240        }
241        let comp: &[i32] = &[4, 2, 5];
242        assert_eq!(&*mutex.lock(), comp);
243    }
244
245    #[test]
246    fn test_mutexguard_sync() {
247        fn sync<T: Sync>(_: T) {}
248
249        let mutex = FairMutex::new(());
250        sync(mutex.lock());
251    }
252
253    #[test]
254    fn test_mutex_debug() {
255        let mutex = FairMutex::new(vec![0u8, 10]);
256
257        assert_eq!(format!("{:?}", mutex), "Mutex { data: [0, 10] }");
258        let _lock = mutex.lock();
259        assert_eq!(format!("{:?}", mutex), "Mutex { data: <locked> }");
260    }
261
262    #[cfg(feature = "serde")]
263    #[test]
264    fn test_serde() {
265        let contents: Vec<u8> = vec![0, 1, 2];
266        let mutex = FairMutex::new(contents.clone());
267
268        let serialized = serialize(&mutex).unwrap();
269        let deserialized: FairMutex<Vec<u8>> = deserialize(&serialized).unwrap();
270
271        assert_eq!(*(mutex.lock()), *(deserialized.lock()));
272        assert_eq!(contents, *(deserialized.lock()));
273    }
274}
parking_lot/fair_mutex.rs

parking_lot/
fair_mutex.rs
