Bringing Rust to Safety-Critical Systems in Space(arxiv.org)
arxiv.org
Bringing Rust to Safety-Critical Systems in Space
https://arxiv.org/abs/2405.18135
22 comments
> Publish a formal language standard first.
You don't need a formal language standard to make Rust usable for safety-critical areas. You do need a formally specified language for the certification, but it doesn't have to be official because heck, you need much more than a mere formally specified language anyway as you've just said. An expanded version of Ferrocene would just fit the bill for example.
I think some people are obsessed about the "formal" specification or the language "standard" because they are misunderstanding their effects. They are absolutely nothing without additional supports. A "formal" specification in this context is not really mathematically formal (like, say, WebAssembly [1]); it merely means that a specification is written in clearly defined terms and logically decomposable statements. A language "standard" is useless if it isn't or can't be enforced, and it doesn't even require the (non-mathematical) formality (though most standard organizations would demand that). As for now, the Rust language specification is "formal" and "standard" enough to be useful for language users, while it remains useful for third-party implementors but they can still find some gaps here and there.
[1] https://webassembly.github.io/spec/core/
You don't need a formal language standard to make Rust usable for safety-critical areas. You do need a formally specified language for the certification, but it doesn't have to be official because heck, you need much more than a mere formally specified language anyway as you've just said. An expanded version of Ferrocene would just fit the bill for example.
I think some people are obsessed about the "formal" specification or the language "standard" because they are misunderstanding their effects. They are absolutely nothing without additional supports. A "formal" specification in this context is not really mathematically formal (like, say, WebAssembly [1]); it merely means that a specification is written in clearly defined terms and logically decomposable statements. A language "standard" is useless if it isn't or can't be enforced, and it doesn't even require the (non-mathematical) formality (though most standard organizations would demand that). As for now, the Rust language specification is "formal" and "standard" enough to be useful for language users, while it remains useful for third-party implementors but they can still find some gaps here and there.
[1] https://webassembly.github.io/spec/core/
The linked spec of wasm is written in English. Can we even say it's "formal"?
Did anyone specify wasm using something like Coq, Lean or Isabelle/HOL?
Did anyone specify wasm using something like Coq, Lean or Isabelle/HOL?
WebAssembly is specified in an English prose and a formal notation, as you can see from, for example, [1]. It is still true that a (mathematically) formal specification can suffer from bugs due to the lack of verification, but it is much easier to verify an already formal specification than a prose specification. It is also possible to turn a prose into a formal specification if the prose in question is already close enough to the formal notation, like ESMeta [2].
[1] https://webassembly.github.io/spec/core/valid/instructions.h...
[2] https://github.com/es-meta/esmeta
[1] https://webassembly.github.io/spec/core/valid/instructions.h...
[2] https://github.com/es-meta/esmeta
I agree that people mis-use the “formal” word when it comes to specifications.
But on this specific question, https://www.cl.cam.ac.uk/~caw77/papers/mechanising-and-verif... is the work I am aware of.
But on this specific question, https://www.cl.cam.ac.uk/~caw77/papers/mechanising-and-verif... is the work I am aware of.
There is a qualified and formally verified compiler called ferrocene [1].
[1]: https://ferrocene.dev/en/
[1]: https://ferrocene.dev/en/
Nit: it's formally specified, not formally verified (which is a really rare and difficult feat in comparison).
Ferrocene notably doesn't support any safety critical CPUs in its current qualification, and those that are experimental only support bare metal. It's exciting, but not at quite the same level of maturity yet.
I'm pretty sure that there are safety critical ARM CPU's and Ferrocene does support ARM.
The R cores are not on the supported target list, they're on the experimental target list. See for yourself: https://public-docs.ferrocene.dev/main/user-manual/targets/i...
I stand corrected. I spoke before I checked my info.
There's also HighTec's compiler that is also ISO 26262 qualified, targeting Infineo's Aurix TriCore:
https://www.infineon.com/cms/en/about-infineon/press/market-...
Not sure tho' what kind of qualification you have to look at in order to fly into space (DO... something?), this is targeting automotive ( and so is Ferrocene's). But a lot of the integrity mechanisms these MCU's have are surely similar between applications.
And... I'll be honest. I really am not convinced by the qualification itself that the quality is necessarily better. It's just that you need it for certain applications.
There's also HighTec's compiler that is also ISO 26262 qualified, targeting Infineo's Aurix TriCore:
https://www.infineon.com/cms/en/about-infineon/press/market-...
Not sure tho' what kind of qualification you have to look at in order to fly into space (DO... something?), this is targeting automotive ( and so is Ferrocene's). But a lot of the integrity mechanisms these MCU's have are surely similar between applications.
And... I'll be honest. I really am not convinced by the qualification itself that the quality is necessarily better. It's just that you need it for certain applications.
>The current state of Rust is a moving target for such tools and design methods. Publish a formal language standard first.
Would it be just easier to use something like OCaml? where it offers the memory safety and a mature language.
Would it be just easier to use something like OCaml? where it offers the memory safety and a mature language.
Garbage collection seems like a significant no-go for many safety critical systems. Other oddities like 31 and 63 bit numbers probably aren't a big deal but are still weird. "Mature" seems like a stretch when OCaml only got proper multi threading 1.5 years ago.
>Publish a formal language standard first.
They tried but they were on the "nightly" branch of the standard and the compiler was on the "release" branch and then the compiler refused to compile the "nightly" standard because "nightly". /s
They tried but they were on the "nightly" branch of the standard and the compiler was on the "release" branch and then the compiler refused to compile the "nightly" standard because "nightly". /s
>Its language characteristics make it trivial to accidentally introduce memory safety issues resulting in undefined behavior or security vulnerabilities.
I worked for a space-related defense contractor. They are using MISRA C, and everything is very thoroughly designed, documented, and tested. I haven’t seen any research into bugs in safety critical systems caused by memory safety issues, but I think that in practice it is very possible to write memory safe C code if you follow industry standards.
I worked for a space-related defense contractor. They are using MISRA C, and everything is very thoroughly designed, documented, and tested. I haven’t seen any research into bugs in safety critical systems caused by memory safety issues, but I think that in practice it is very possible to write memory safe C code if you follow industry standards.
Skimming MISRA C, it seems to be a collection of either extremely limiting or ineffectual directives. An example of each:
> Dir 4.12 Dynamic memory allocation shall not be used
> Rule 2.2 There shall be no dead code
https://github.com/sakura1083841400/MISRA-C/blob/main/MISRA%...
This provides nothing approaching the guarantees of safe Rust, as far as I can tell.
> Dir 4.12 Dynamic memory allocation shall not be used
> Rule 2.2 There shall be no dead code
https://github.com/sakura1083841400/MISRA-C/blob/main/MISRA%...
This provides nothing approaching the guarantees of safe Rust, as far as I can tell.
Not sure what you’re getting at. We take those rules and use a static analysis tool (LDRA) to prove they are met. You can write unsafe code in any language.
I wouldn't dismiss MISRA C so quickly as its guidelines may prove useful when deriving best practices for unsafe Rust.
Just curious: do you see any issues with concurrency or anything else rust would help with, or is it domain knowledge related issues?
I am answering this comment from the audience at https://oxidizeconf.com/
There is a growing number of people and organizations that do see Rust as bringing advantages in this area.
That said, your parent isn’t wrong that the safety critical space has developed numerous ways of mitigating the risk of writing software, no matter what language it’s written in. But that doesn’t mean that language is irrelevant.
There is a growing number of people and organizations that do see Rust as bringing advantages in this area.
That said, your parent isn’t wrong that the safety critical space has developed numerous ways of mitigating the risk of writing software, no matter what language it’s written in. But that doesn’t mean that language is irrelevant.
Very cool project! I am also working on a similar topic as part of my PhD thesis.
Why wouldn't they use Ada/SPARK2014? They have a published language standard, verification tools, and 25 years of applied use for mission-critical, high-integrity software.
The current state of Rust is a moving target for such tools and design methods. Publish a formal language standard first.
https://www.adacore.com/about-spark