Skip to content

VoodooChild99/perry-experiments

Repository files navigation

Perry Experiments

This directory contains the data required to reproduce the experiments described in our USENIX 2024 paper: A Friend's Eye is A Good Mirror: Synthesizing MCU Peripheral Models from Peripheral Drivers.

For driver source code we used in our experiments, please refer to this repo.

Requirements

Hardware Dependencies

  • Processor: We used two Intel Xeon E5-2620 v2 CPUs (12 cores, 24 threads). Comparable hardware may also suffice.
  • Memory: At least 64GB of RAM
  • Bluetooth: The machine must be equipped with a Bluetooth device to replicate 04-cve-reproducing

Software Dependencies

  • OS: We used Ubuntu 20.04. Other systems are not tested.
  • Misc: git and Docker

Preparations

It is assumed that you have built Perry with Docker as described here. All experiments should be conducted within the built container (execute /perry/run_docker.sh to launch a shell). This repository will be located under the /root/perry-experiments directory within the container.

Directory Layout

Directories/Files Experiment
01-efficiency Model synthesis time consumption as described in Section 5.2 ("Efficiency")
02-consistency P2IM unit tests passing rate as described in Section 5.2 ("Consistency" and "Scalability")
03-universality Emulating various firmware samples as described in Section 5.2 ("Universality" and "Scalability")
spec-violation-bugs.md The specification violation bugs we found in drivers as described in Section 6.1
04-cve-reproducing Reproducing firmware CVEs as described in Section 6.2 ("CS-II")
05-liteos-fuzzing Fuzzing LiteOS as described in Section 6.2 ("CS-III")
pre-generated-models Hardware models generated by Perry

Running the Experiments

Please refer to the README files in 01-efficiency, 02-consistency, 03-universality, 04-cve-reproducing and 05-liteos-fuzzing for detailed instructions.

Note that experiment 02-consistency and 05-liteos-fuzzing depends on hardware models generated in 01-efficiency. Hardware model generation takes ~20 hours. To facilitate reproduction, we provide pre-generated hardware models under the pre-generated-models folder.

Badges

Releases

No releases published

Packages

No packages published

Languages