wait_on_address

Cross platform atomic wait and wake (aka futex) functionality. This crate is a fork of atomic-wait, and extends the original code with the following functionality:

• Support for AtomicI32, AtomicI64, and AtomicU64
• Support for waiting with a timeout
• Support for wasm32 on nightly using std::arch
• Polyfill for all other platforms

Natively-supported platforms:

• Windows 8+, Windows Server 2012+
• macOS 14.4+, iOS 17.4+, watchOS 10.4+
• Linux 2.6.22+ (using fallback for 64-bit futexes)
• wasm32
• All other platforms with std support (using fallback)

Usage

use std::{sync::atomic::AtomicU64, time::Duration};
use wait_on_address::AtomicWait;

let a = AtomicU64::new(0);

a.wait(1); // If the value is 1, wait.

a.wait_timeout(2, Duration::from_millis(100));  // If the value is 2, wait at most 100 milliseconds

a.notify_one(); // Wake one waiting thread.

a.notify_all(); // Wake all waiting threads.

Implementation

On Linux, this uses the SYS_futex syscall.

On FreeBSD, this uses the _umtx_op syscall.

On Windows, this uses the WaitOnAddress and WakeByAddress APIs.

On macOS (and iOS and watchOS), this uses the os_sync_wait_on_address and os_sync_wake_by_address APIs.

On wasm32 with nightly, this uses memory_atomic_wait32, memory_atomic_wait64, and memory_atomic_notify instructions.

All other platforms with std support fall back to a fixed-size hashmap of Condvars, similar to libstdc++'s implementation for std::atomic<T>.