CELLIFT: Leveraging Cells for Scalable and Precise Dynamic Information Flow Tracking in RTL

Abstract

Dynamic Information Flow Tracking (dynamic IFT) is a well-known technique with many security applications such as analyzing the behavior of a system given an input and detecting security violations. While there are many widely used open dynamic IFT solutions that scale to large software, the same level of support is unfortunately lacking for hardware. This gap is becoming more pronounced with the increasing complexity of open-source hardware and the plethora of recent hardware attacks. We introduce CellIFT, a new design point in the space of dynamic IFT for hardware. CellIFT leverages the logical macrocell abstraction (e.g., an adder) to achieve scalability, precision and completeness when instrumenting a given Register Transfer Level (RTL) hardware design. Cell-level dynamic IFT does not suffer from the scalability problems that are inherent to lower levels of abstraction such as gates, yet it achieves completeness given the limited number of cell types. We show the versatility of CellIFT by instrumenting five distinct RISC-V designs, one of which is a complete SoC. The only existing complete solution already fails to instrument two of these designs. Our extensive evaluation using microbenchmarks and standard RISC-V benchmarks on the instrumented designs shows that CellIFT is 21× to 61× faster than the state of the art in terms of simulation runtime without losing precision. We further show-case concrete applications of CellIFT in four scenarios by detecting: 1) sources of microarchitectural information leakage, 2) microarchitectural bugs such as Meltdown, 3) speculative vulnerabilities such as Spectre-BCB, and 4) SoC-wide architectural design flaws. We release CellIFT as open source to enable RTL-level security research for the wider community.