Rust Atomics and Locks

Chapter 9. Building Our Own Locks

In this chapter, we’ll build our own mutex, condition variable, and reader-writer lock. For each of them, we’ll start with a very basic version, and then extend it to make it more efficient.

Since we’re not going to use the lock types from the standard libary (which would be cheating), we’re going to have to use the tools from Chapter 8 to be able to make threads wait without busy-looping. However, as we saw in that chapter, the available tools the operating system provides vary a lot per platform, making it hard to build something that works cross-platform.

Fortunately, most modern operating systems support futex-like functionality, or at least the wake and wait operations. As we saw in Chapter 8, Linux has supported them since 2003 with the futex syscall, Windows since 2012 with the WaitOnAddress family of functions, FreeBSD since 2016 as part of the _umtx_op syscall, and so on.

The most notable exception is macOS. While its kernel does support these operations, it is not exposed through any stable, publicly usable, C function that we can use. However, macOS does ship with a recent version of libc++, an implementation of the C++ standard library. This library includes support for C++20, which is the version of C++ that comes with built-in support for very basic atomic wait and wake operations (like std::atomic<T>::wait()). While it’s somewhat tricky to make use of that from Rust for a variety of reasons, it is certainly possible, giving us ...