11 releases

0.3.1 Oct 9, 2024
0.3.0 Aug 26, 2024
0.3.0-alpha-1 Jun 28, 2024
0.2.0 Mar 9, 2024
0.1.0-alpha.3 Oct 16, 2023

#188 in Embedded development


Used in machine-check-avr

MIT/Apache

28KB
401 lines

Machine-check: a formal verification tool for digital systems

Machine-check is a tool for formal verification of digital system properties, currently in experimental stage.

Unlike classic software testing, which can find bugs but not prove their absence, formal verification can prove that specified undesirable behaviours can never occur in the system. Unfortunately, formal verification requires complicated reasoning (e.g. with intervals instead of just numbers) and advanced techniques are necessary for its automation. Machine-check aims to provide these techniques, but shield the user from them as much as possible.

The main intended use-case of machine-check is formal verification of machine-code programs on simple microcontrollers, but the approach is highly general, allowing verification of arbitrary digital systems as long as they are described in a subset of valid Rust code that machine-check understands.

Examples

A very simple example of a system verifiable by machine-check is counter, a simple finite-state machine which contains an eight-bit value, which is initialized to zero on initialization and then is incremented if the increment single-bit input is 1. If the value reaches 157, it is immediately zeroed again. The system is very simple, so it is complicated a little by a large unused bitvector array, which would make simple kinds of automated verification impossible.

The magic of machine-check is unlocked by the machine_description macro, which adds verification analogues to the code it is applied to, allowing simply running machine-check by calling [run] in the main function after constructing the system.

If you put the counter inside your own Rust crate (with machine-check as a dependency), you can verify that the counter is always lesser than 157 in each system state, using a specification property based on Computation Tree Logic. Let's use machine-check to prove the property:

$ cargo run -- --property "AG[unsigned_lt(value,157)]"

[2024-03-07T22:44:44Z INFO  machine_check_exec] Starting verification.
[2024-03-07T22:44:44Z INFO  machine_check_exec] Verification ended.
{"result":{"Ok":true},"stats":{"num_states":178,"num_refinements":309}}

(Note that the final specification and output formats are still under construction and will be nicer than this.)

On the other hand, machine-check tells us that the counter value is NOT always lesser than 156:

$ cargo run -- --property "AG[unsigned_lt(value,156)]"

$ cargo run -- --property "AG[unsigned_lt(value,156)]"
[2024-03-07T22:45:47Z INFO  machine_check_exec] Starting verification.
[2024-03-07T22:45:47Z INFO  machine_check_exec] Verification ended.
{"result":{"Ok":false},"stats":{"num_states":178,"num_refinements":308}}

You can use the "-v" command-line parameter to show the final abstract state space, although interpreting it requires some further knowledge.

Inherent panics

It is also possible to detect system panics, which is useful e.g. for machine-code systems, which take the machine-code file from command line and should detect that it a reserved instruction can be executed during the course of the program. A simple example is in the conditional_panic example, which should panic with message "Example panic 2" if the input is equal to 1. You can copy it into your crate, then run it with parameter "--inherent", which signifies that you are only interested about the inherent panics of the system:

$ cargo run -- --inherent
[2024-03-07T22:59:26Z INFO  machine_check_exec] Starting verification.
[2024-03-07T22:59:26Z ERROR machine_check_exec] Verification failed.
{"result":{"Err":{"InherentPanic":"Example panic 2"}},"stats":{"num_states":3,"num_refinements":8}}

Currently, presence of inherent panic precludes verification of a property, as it is a more pressing issue to fix.

Machine-code verification

There is also an example of an extremely simplified RISC microcontroller in simple_risc, showcasing the bitmask_switch macro that can be used for elegant transcription of microcontroller behaviour depending on instruction opcodes.

A more proper implementation of an actual microcontroller is present in the crate machine-check-avr, allowing verification of some simple machine-code programs for the AVR ATmega328P microcontroller (notably used in Arduino Uno R3).

Current status

Machine-check is still in experimental phase, with limitations in user experience and verification power.

The machine_description macro, in particular, is currently very finicky and errors produced may or may not be useful. If you want to try writing your own system, proceed step by step, slowly adding and modifying pieces of example code. Temporarily commenting out the macro may also reveal an underlying Rust error with a more sensible error message.

Further notes

Unlike some other formal verification tools, machine-check is designed be sound and complete. You should either get an error or a correct true/false result in finite (but practically unbounded) time. Of course, there may be bugs or design oversights.

Changelog

  • 0.3.1: Continuing a refinement until the state space changes. This improves performance a bit in some scenarios.
  • 0.3.0: Soundness fixes, optimisation, refinement choice tweaks for reasonable verification of machine-code systems.
  • 0.2.0: Significant rewrite, arbitrary finite-state systems now can be described as finite-state machines in Rust code. Conditional branches are supported.
  • 0.1.0: Initial version, only verification of [Btor2] (https://link.springer.com/chapter/10.1007/978-3-319-96145-3_32) hardware systems supported through translation to Rust code.

License

Licensed under either of Apache License, Version 2.0 or MIT license at your option. Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in this crate by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.

Dependencies

~8–10MB
~181K SLoC