Crate kernel

source ·
Expand description

Core Tock Kernel

The kernel crate implements the core features of Tock as well as shared code that many chips, capsules, and boards use. It also holds the Hardware Interface Layer (HIL) definitions.

Most unsafe code is in this kernel crate.

Core Kernel Visibility

As the root crate in the Tock operating system, this crate serves multiple purposes:

  1. It includes the logic for the core kernel, including process management, grants, scheduling, etc.

  2. It includes important interfaces for hardware and other device abstractions. These are generally in the HIL and platform folders.

  3. It includes utility functions used elsewhere in the kernel, generally by multiple different crates such that it makes sense to have shared implementations in the core kernel crate.

Because of these different features of the core kernel, managing visibility of the various objects and functions is a bit tricky. In general, the kernel crate only exposes what it absolutely needs to. However, there are three cases where resources in this crate must be exposed.

  1. The shared utility functions and structs must be exposed. These are marked pub and are used by many other kernel crates.

    Some utility objects and abstractions, however, expose memory unsafe behavior. These are marked as unsafe, and require an unsafe block to use them. One example of this is StaticRef which is used for accessing memory-mapped I/O registers. Since accessing the addresses through just a memory address is potentially very unsafe, instantiating a StaticRef requires an unsafe block.

  2. The core kernel types generally have to be exposed as other layers of the OS need to use them. However, generally only a very small interface is exposed, and using that interface cannot compromise the overall system or the core kernel. These functions are also marked pub. For example, the ProcessBuffer abstraction must be exposed to capsules to use shared memory between a process and the kernel. However, the constructor is not public, and the API exposed to capsules is very limited and confined by the Rust type system. The constructor and other sensitive interfaces are restricted to use only inside the kernel crate and are marked pub(crate).

    In some cases, more sensitive core kernel interfaces must be exposed. For example, the kernel exposes a function for starting the main scheduling loop in the kernel. Since board crates must be able to start this loop after all initialization is finished, the kernel loop function must be exposed and marked pub. However, this interface is not generally safe to use, since starting the loop a second time would compromise the stability of the overall system. It’s also not necessarily memory unsafe to call the start loop function again, so we do not mark it as unsafe. Instead, we require that the caller hold a Capability to call the public but sensitive functions. More information is in capabilities.rs. This allows the kernel crate to still expose functions as public while restricting their use. Another example of this is the Grant constructor, which must be called outside of the core kernel, but should not be called except during the board setup.

  3. Certain internal core kernel interfaces must also be exposed. These are needed for extensions of the core kernel that happen to be implemented in crates outside of the kernel crate. For example, additional implementations of Process may live outside of the kernel crate. To successfully implement a new Process requires access to certain in-core-kernel APIs, and these must be marked pub so that outside crates can access them.

    These interfaces are highly sensitive, so again we require the caller hold a Capability to call them. This helps restrict their use and makes it very clear that calling them requires special permissions. Additionally, to differentiate these interfaces, which are for external extensions of core kernel functionality, from the other public but sensitive interfaces (item 2 above), we append the name _external to the function name.

    One note is that there are currently very few extensions to the core kernel that live outside of the kernel crate. That means we have not necessarily created _extern functions for all the interfaces needed for this use case. It is likely we will have to create new interfaces as new use cases are discovered.

Re-exports

  • pub use crate::errorcode::ErrorCode;
  • pub use crate::kernel::Kernel;
  • pub use crate::process::ProcessId;
  • pub use crate::scheduler::Scheduler;

Modules

  • Special restricted capabilities.
  • Data structures.
  • Components extend the functionality of the Tock kernel through a simple factory method interface.
  • Support for in-kernel debugging.
  • Hardware-independent kernel interface for deferred calls
  • Standard errors in Tock.
  • Support for processes granting memory from their allocations to the kernel.
  • Public traits for interfaces between Tock components.
  • Mechanism for inspecting the status of the kernel.
  • Inter-process communication mechanism for Tock.
  • Traits for implementing various layers and components in Tock.
  • Types for Tock-compatible processes.
  • Traits and types for application credentials checkers, used to decide whether an application can be loaded.
  • Data structures for passing application memory to the kernel.
  • Interface for Tock kernel schedulers.
  • Mechanism for managing storage read & write permissions.
  • Mechanisms for handling and defining system calls.
  • Data structure for storing an upcall from the kernel to a process.
  • Utility functions and macros provided by the kernel crate.

Macros

  • Count the number of passed expressions.
  • Create an object with the given capability.
  • In-kernel println() debugging.
  • This macro prints a new line to an internal ring buffer, the contents of which are only flushed with debug_flush_queue! and in the panic handler.
  • Prints out the expression and its location, then returns it.
  • This macro flushes the contents of the debug queue into the regular debug output.
  • In-kernel gpio debugging, accepts any GPIO HIL method
  • In-kernel println() debugging that can take a process slice.
  • In-kernel println() debugging with filename and line numbers.
  • Allocates a statically-sized global region of memory for data structures but does not initialize the memory. Checks that the buffer is not aliased and is only used once.
  • Allocates a statically-sized global array of memory and initializes the memory for a particular data structure.
  • Allocates space in the kernel image for on-chip non-volatile storage.

Constants