Crate maitake

maitake

🎶🍄 “Dancing mushroom” — an async runtime construction kit.

✨ as seen in the rust compiler test suite!

what is it?

This library is a collection of modular components for building a Rust async runtime based on core::task and core::future, with a focus on supporting #![no_std] projects.

Unlike other async runtime implementations, maitake does not provide a complete, fully-functional runtime implementation. Instead, it provides reusable implementations of common functionality, including a task system, scheduler, a timer wheel, and synchronization primitives. These components may be combined with other runtime services, such as timers and I/O resources, to produce a complete, application-specific async runtime.

maitake was initially designed for use in the mycelium and mnemOS operating systems, but may be useful for other projects as well.

Note

This is a hobby project. I’m working on it in my spare time, for my own personal use. I’m very happy to share it with the broader Rust community, and contributions and bug reports are always welcome. However, please remember that I’m working on this library for fun, and if it stops being fun…well, you get the idea.

Anyway, feel free to use and enjoy this crate, and to contribute back as much as you want to!

a tour of maitake

maitake currently provides the following major API components:

• maitake::task: the maitake task system. This module contains the Task type, representing an asynchronous task (a Future that can be spawned on the runtime), and the TaskRef type, a reference-counted, type-erased pointer to a spawned Task.

  Additionally, it also contains other utility types for working with tasks. These include the JoinHandle type, which can be used to await the output of a task once it has been spawned, and the task::Builder type, for configuring a task prior to spawning it.

• maitake::scheduler: schedulers for executing tasks. In order to actually execute asynchronous tasks, one or more schedulers is required. This module contains the Scheduler and StaticScheduler types, which implement task schedulers, and utilities for constructing and using schedulers.

• maitake::time: timers and futures for tracking time. This module contains tools for waiting for time-based events in asynchronous systems. It provides the Sleep type, a Future which completes after a specified duration, and the Timeout type, which wraps another Future and cancels it if it runs for longer than a specified duration without completing.

  In order to use these futures, a system must have a timer. The maitake::time module therefore provides the Timer type, a hierarchical timer wheel which can track and notify a large number of time-based futures efficiently. A Timer must be driven by a hardware time source, such as an interrupt or timestamp counter.

• maitake::sync: asynchronous synchronization primitives. This module provides asynchronous implementations of common synchronization primitives, including a Mutex, RwLock, and Semaphore. Additionally, it provides lower-level synchronization types which may be useful when implementing custom synchronization strategies.

• maitake::future: utility futures. This module provides general-purpose utility Future types that may be used without the Rust standard library.

usage considerations

maitake is intended primarily for use in bare-metal projects, such as operating systems, operating system components, and embedded systems. These bare-metal systems typically do not use the Rust standard library, so maitake supports #![no_std] by default, and the use of liballoc is feature-flagged for systems where liballoc is unavailable.

This intended use case has some important implications:

maitake is not a complete asynchronous runtime

This is in contrast to other async runtimes, like tokio, async-std, and glommio, which provide everything a userspace application needs to run async tasks and perform asynchronous IO, with lower-level implementation details encapsulated behind the runtime’s API. In the bare-metal systems maitake is intended for use in, however, it is often necessary to have more direct control over lower-level implementation details of the runtime.

For example: in an asynchronous runtime, tasks must be stored in non-stack memory. Runtimes like tokio and async-std use the standard library’s allocator and Box type to allocate tasks on the heap. In bare-metal systems, though, liballoc’s heap allocator may not be available. Such a system may have no ability to perform dynamic heap allocations, or may implement its own allocator which may not be compatible with liballoc.

maitake is designed to still be usable in those cases — even a system which cannot dynamically allocate memory could use maitake in order to create and schedule a fixed set of tasks that are stored in 'statics, essentially allocating tasks at compile-time. Therefore, maitake provides an interface for overriding the memory container in which tasks are stored. In order to provide such an interface, however, maitake must expose the in-memory representation of spawned tasks, which other runtimes typically do not make part of their public APIs.

maitake does not support unwinding

Rust supports multiple modes of handling panics: panic="abort" and panic="unwind". When a program is compiled with panic="unwind", panics are handled by unwinding the stack of the panicking thread. This allows the use of APIs like catch_unwind, which allows panics to be handled without terminating the entire program. On the other hand, compiling with panic="abort" means that all panics immediately terminate the program.

Bare-metal systems typically do not use stack unwinding. For programs which use the Rust standard library, support for unwinding is provided by std. However, in bare-metal, #![no_std] systems, it is necessary for the system to implement its own unwinding system. Therefore, maitake does not support unwinding.

This is important to note, because supporting unwinding imposes additional safety considerations. In order to safely support unwinding, many parts of maitake, such as runtime internals and synchronization primitives, would have to take extra steps to ensure that they cannot be left in an invalid state during unwinding. Ensuring unwind-safety would require the use of standard library APIs that are not available without std, so maitake does not ensure unwind-safety.

This means that maitake should not be used in programs compiled with panic="unwind". Typically, no bare-metal program will fall into this category, but if you are using maitake in a project which uses std, it is necessary to explicitly disable unwinding in that project’s Cargo.toml.

platform support

In general, maitake is a platform-agnostic library. It does not interact directly with the underlying hardware, or use platform-specific features (with one small exception). Instead, maitake provides portable implementations of core runtime components. In some cases, such as the timer wheel, downstream code must integrate maitake’s APIs with hardware-specific code for in order to use them effectively.

support for atomic operations

However, one aspect of maitake’s implementation may differ slightly across different target architectures: maitake relies on atomic operations integers. Sometimes, atomic operations on integers of specific widths are needed (e.g., AtomicU64), which may not be available on all architectures.

In order to work on architectures which lack atomic operations on 64-bit integers, maitake uses the portable-atomic crate by Taiki Endo. This crate crate polyfills atomic operations on integers larger than the platform’s pointer width, when these are not supported in hardware.

In most cases, users of maitake don’t need to be aware of maitake’s use of portable-atomic. If compiling maitake for a target architecture that has native support for 64-bit atomic operations (such as x86_64 or aarch64), the native atomics are used automatically. Similarly, if compiling maitake for any target that has atomic compare-and-swap operations on any size integer, but lacks 64-bit atomics (i.e., 32-bit x86 targets like i686, or 32-bit ARM targets with atomic operations), the portable-atomic polyfill is used automatically. Finally, when compiling for target architectures which lack atomic operations because they are always single-core, such as MSP430 or AVR microcontrollers, portable-atomic simply uses unsynchronized operations with interrupts temporarily disabled.

The only case where the user must be aware of portable-atomic is when compiling for targets which lack atomic operations but are not guaranteed to always be single-core. This includes ARMv6-M (thumbv6m), pre-v6 ARM (e.g., thumbv4t, thumbv5te), and RISC-V targets without the A extension. On these architectures, the user must manually enable the RUSTFLAGS configuration --cfg portable_atomic_unsafe_assume_single_core if (and only if) the specific target hardware is known to be single-core. Enabling this cfg is unsafe, as it will cause unsound behavior on multi-core systems using these architectures.

Additional configurations for some single-core systems, which determine the specific sets of interrupts that portable-atomic will disable when entering a critical section, are described here.

features

The following features are available (this list is incomplete; you can help by expanding it.)

Feature	Default	Explanation
alloc	true	Enables liballoc dependency
no-cache-pad	false	Inhibits cache padding for the CachePadded struct. When this feature is NOT enabled, the size will be determined based on target platform.
tracing-01	false	Enables support for v0.1.x of tracing (the current release version). Requires liballoc.
tracing-02	false	Enables support for the upcoming v0.2 of tracing (via a Git dependency).
core-error	false	Enables implementations of the core::error::Error trait for maitake’s error types. Requires a nightly Rust toolchain.

Modules

• future
• scheduler
  Schedulers for executing tasks.
• sync
  Asynchronous synchronization primitives
• task
  The maitake task system.
• time
  Utilities for tracking time and constructing system timers.

Macros

• new_static_scheduler
  Safely constructs a new StaticScheduler instance in a static initializer.