README.md
1# spin-rs
2
3[](https://crates.io/crates/spin)
4[](https://docs.rs/spin/)
5[](https://travis-ci.org/mvdnes/spin-rs)
6
7Spin-based synchronization primitives.
8
9This crate provides [spin-based](https://en.wikipedia.org/wiki/Spinlock)
10versions of the primitives in `std::sync`. Because synchronization is done
11through spinning, the primitives are suitable for use in `no_std` environments.
12
13Before deciding to use `spin`, we recommend reading
14[this superb blog post](https://matklad.github.io/2020/01/02/spinlocks-considered-harmful.html)
15by [@matklad](https://github.com/matklad/) that discusses the pros and cons of
16spinlocks. If you have access to `std`, it's likely that the primitives in
17`std::sync` will serve you better except in very specific circumstances.
18
19## Features
20
21- `Mutex`, `RwLock`, `Once`, `Lazy` and `Barrier` equivalents
22- Support for `no_std` environments
23- [`lock_api`](https://crates.io/crates/lock_api) compatibility
24- Upgradeable `RwLock` guards
25- Guards can be sent and shared between threads
26- Guard leaking
27- Ticket locks
28- Different strategies for dealing with contention
29
30## Usage
31
32Include the following under the `[dependencies]` section in your `Cargo.toml` file.
33
34```toml
35spin = "x.y"
36```
37
38## Example
39
40When calling `lock` on a `Mutex` you will get a guard value that provides access
41to the data. When this guard is dropped, the mutex will become available again.
42
43```rust
44extern crate spin;
45use std::{sync::Arc, thread};
46
47fn main() {
48 let counter = Arc::new(spin::Mutex::new(0));
49
50 let thread = thread::spawn({
51 let counter = counter.clone();
52 move || {
53 for _ in 0..100 {
54 *counter.lock() += 1;
55 }
56 }
57 });
58
59 for _ in 0..100 {
60 *counter.lock() += 1;
61 }
62
63 thread.join().unwrap();
64
65 assert_eq!(*counter.lock(), 200);
66}
67```
68
69## Feature flags
70
71The crate comes with a few feature flags that you may wish to use.
72
73- `mutex` enables the `Mutex` type.
74
75- `spin_mutex` enables the `SpinMutex` type.
76
77- `ticket_mutex` enables the `TicketMutex` type.
78
79- `use_ticket_mutex` switches to a ticket lock for the implementation of `Mutex`. This
80 is recommended only on targets for which ordinary spinning locks perform very badly
81 because it will change the implementation used by other crates that depend on `spin`.
82
83- `rwlock` enables the `RwLock` type.
84
85- `once` enables the `Once` type.
86
87- `lazy` enables the `Lazy` type.
88
89- `barrier` enables the `Barrier` type.
90
91- `lock_api` enables support for [`lock_api`](https://crates.io/crates/lock_api)
92
93- `std` enables support for thread yielding instead of spinning.
94
95- `portable_atomic` enables usage of the `portable-atomic` crate
96 to support platforms without native atomic operations (Cortex-M0, etc.).
97 The `portable_atomic_unsafe_assume_single_core` cfg or `critical-section` feature
98 of `portable-atomic` crate must also be set by the final binary crate.
99
100 When using the cfg, this can be done by adapting the following snippet to the `.cargo/config` file:
101 ```
102 [target.<target>]
103 rustflags = [ "--cfg", "portable_atomic_unsafe_assume_single_core" ]
104 ```
105 Note that this cfg is unsafe by nature, and enabling it for multicore systems is unsound.
106
107 When using the `critical-section` feature, you need to implement the critical-section
108 implementation that sound for your system by implementing an unsafe trait.
109 See [the documentation for the `portable-atomic` crate](https://docs.rs/portable-atomic/latest/portable_atomic/#optional-cfg)
110 for more information.
111
112## Remarks
113
114It is often desirable to have a lock shared between threads. Wrapping the lock in an
115`std::sync::Arc` is route through which this might be achieved.
116
117Locks provide zero-overhead access to their data when accessed through a mutable
118reference by using their `get_mut` methods.
119
120The behaviour of these lock is similar to their namesakes in `std::sync`. they
121differ on the following:
122
123- Locks will not be poisoned in case of failure.
124- Threads will not yield to the OS scheduler when encounter a lock that cannot be
125 accessed. Instead, they will 'spin' in a busy loop until the lock becomes available.
126
127Many of the feature flags listed above are enabled by default. If you're writing a
128library, we recommend disabling those that you don't use to avoid increasing compilation
129time for your crate's users. You can do this like so:
130
131```
132[dependencies]
133spin = { version = "x.y", default-features = false, features = [...] }
134```
135
136## Minimum Safe Rust Version (MSRV)
137
138This crate is guaranteed to compile on a Minimum Safe Rust Version (MSRV) of 1.38.0 and above.
139This version will not be changed without a minor version bump.
140
141## License
142
143`spin` is distributed under the MIT License, (See `LICENSE`).
144
){kind=link}
){kind=link}
){kind=link}