Branch predictor (BP) is one of the key performance improvement mechanisms in today's processors. Recent studies demonstrate that it can be used to initiate powerful attacks such as side-channel and speculative execution-based attacks. These attacks allow adversaries to steal sensitive data and compromise computer systems. This project investigates security threats introduced by existing BP designs and develops new safe designs to stop BP-related attacks without significantly degrading the performance.
First, the project addresses the problem of sensitive data leakage through shared BP data structures. It develops a safe sharing mechanism which permits BP structure sharing while simultaneously preventing dangerous collisions. Second, the project addresses the speculative execution threats caused by incorrectly predicted branch instructions. This is achieved by redesigning speculative execution components of the processor by adding restrictions on speculative instruction execution and caching critical data to prevent drastic performance losses. Third, the project develops a novel evaluation framework that is used to evaluate and compare security and performance properties of new and existing branch predictor designs.
Hardware security vulnerabilities affect billions of computers worldwide. This project intends to improve the security of computer hardware and disallow both known and future attacks by designing safe branch predictors. As a result, security of future computers will be improved, making a positive impact on society. Discoveries from this project are used in curriculum development. The project actively involves graduate students into research.
Data, code and other research artifacts generated by this project will be stored locally at William and Mary machines and backed up to an in-house network storage system. Tools and data samples will be available at the project website: www.cs.wm.edu/~dmitry/safebp/. All data generated during this project will be kept for the duration of the project and at least two additional years.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
