Isla

Isla is a symbolic execution engine for Sail, and a tool (sometimes known more specifically as isla-axiomatic) that uses that to evaluate the relaxed-memory behavior of instruction set architectures (ISAs) specified in Sail, including Armv8-A and RISC-V, with respect to arbitrary axiomatic memory models specified in a subset of the cat language used by the herd7 tools. For example, for a classic message-passing test on Armv8-A, Isla finds the following candidate execution satisfying the final condition of the test, with the instruction behaviour taken from symbolic evaluation of the full Armv8-A ISA definition.

There is an online web interface for isla-axiomatic here:

https://isla-axiomatic.cl.cam.ac.uk

Isla has also been used for test generation, generating simplified semantics (summaries) for concrete opcodes, and there are many other possible use cases.

Build

Get Rust and its dependencies, probably the easiest is with rustup.

Then build isla and friends, currently tested with (stable) Rust 1.70:

cargo build --release

Z3

By default we require that z3 is available as a shared library.

If running isla-axiomatic, z3 must also be available on the PATH.

It should be available in the usual places, currently tested with Z3 4.12.6:

apt install libz3-dev  # on Ubuntu 20.04 LTS and later
opam install z3 # alterantively, opam usually has an up-to-date version

However the version of z3 that is available in older Ubuntu LTS versions (and likely other linux distros) is quite old, so you may experience link errors in that case. The build.rs script is configured so it can use a libz3.so shared library placed in the root directory of this repository. If this is done then LD_LIBRARY_PATH must also be set when executing so that the more recent z3 library is used. In those cases a version of z3 can be obtained from the sources https://github.com/Z3Prover/z3.

Model snapshots

Isla executes IR produced by Sail. To avoid having to generate this IR, there are pre-compiled snapshots of our ISA models available in the following repository:

https://github.com/rems-project/isla-snapshots

To generate this IR in the correct format a tool is available in the isla-sail directory. Building this requires various arcane OCaml incantations, but mostly one can follow the Sail install guide here, followed by the instructions here. It will only work with the latest HEAD of the sail2branch in the Sail repository.

Litmus test format conversion

For litmus tests in the .litmus format used by herd7 and rmem there is another OCaml tool based on parsing code from herd7 in the isla-litmus directory, which translates that format into a simple TOML representation. This OCaml program is standalone and does not depend on any libraries, and should build with dune >= 1.2.

Example

After compiling Isla, to compute the footprint of an add instruction using the ARM 9.4 snapshot above, the following command can be used:

target/release/isla-footprint -A armv9p4.ir -C configs/armv9p4.toml -i "add x0, x1, #3" -s

The arguments are the compiled Sail model, a configuration file controlling the initial state and other options, and the instruction we want to run. The -s flag performs some basic dead-code elimination to simplify the generated footprint. We get a trace of the instruction as a mix of SMTLIB definitions of the semantics, interspersed with statements describing the input and outputs of the instruction, here read-reg and write-reg.

  (declare-const v3371 (_ BitVec 64))
  (read-reg |R1| nil v3371)
  (define-const v3457 (bvadd ((_ extract 63 0) ((_ zero_extend 64) v3371)) #x0000000000000003))
  (write-reg |R0| nil v3457))

Example: Running isla-axiomatic

To run a relaxed litmus test, first ensure an up-to-date snapshot is obtained from isla-snapshots

A quick and simple test is just to run the Arm MP+dmb.sy+ctrl litmus test, against the same model the web interface uses:

target/release/isla-axiomatic -A /path/to/isla-snapshots/armv8p5.ir -C configs/armv8p5.toml -m web/client/dist/aarch64.cat web/client/dist/aarch64/MP+dmb.sy+ctrl.toml

For more information, see the full documentation.

Manual

There is a guide to the various Isla command line options and features here.

The isla-axiomatic tool has a seperate manual here, and a guide to its support for virtual memory and address translation here.

Project structure

• isla-lib Is a Rust library which contains the core symbolic execution engine and an API to interact with it.

• isla-axiomatic Contains rust code to handle various aspects which are specific to checking axiomatic concurrency models on top of isla-lib, such as parsing litmus tests, analysing instruction footprints, and defining a high-level interface to run litmus tests.

• isla-cat Is a translator from (a fragment of) the cat memory models used by herdtools into SMTLIB definitions. It has its own README here.

• isla-litmus Is an (optional) OCaml utility that maps the .litmus files that herdtools uses into a format we can read.

• isla-sail Is an (optional) OCaml utility that maps Sail specifications into the IR we can symbolically execute.

• web Contains a server and client for a web interface to the axiomatic concurrency tool

• src Defines multiple small executable utilities based on isla-lib

People

• Alasdair Armstrong (University of Cambridge)
• Brian Campbell (University of Edinburgh)
• Ben Simner (University of Cambridge)
• Thibaut Perami (University of Cambridge)
• Peter Sewell (University of Cambridge)

Funding

This software was developed by the University of Cambridge Computer Laboratory (Department of Computer Science and Technology), in part under DARPA/AFRL contract FA8650-18-C-7809 ("CIFV"), in part funded by EPSRC Programme Grant EP/K008528/1 "REMS: Rigorous Engineering for Mainstream Systems", in part funded from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 789108, "ELVER"), in part supported by the UK Government Industrial Strategy Challenge Fund (ISCF) under the Digital Security by Design (DSbD) Programme, to deliver a DSbDtech enabled digital platform (grant 105694), in part funded by an Arm iCASE doctoral studentship (18000005, Simner), and in part funded by Google.