← Lab /Security & privacy

SecDMP eliminates DMP side-channel leaks with minimal overhead

Security & privacy Defense · Prototype · arXiv Threat model

The claim

SecDMP prevents all data memory-dependent prefetcher leakage while preserving performance for public data

Jens Sprengers, Marton Bognar, Lesly-Ann Daniel, Frank Piessens · arXiv Paper PDF

Published as: “SecDMP: Provably Secure Data Memory-Dependent Prefetching (Extended Version)”

So what

SecDMP enables safe use of performance-boosting DMPs in constant-time code without sacrificing security or incurring noticeable overhead.

The five-minute version

SecDMP eliminates all data-memory-dependent prefetcher (DMP) leakage while adding only about 0.36% average performance overhead and less than 0.5% chip-area increase. The authors prove formally that no secret information leaves the microarchitecture beyond a baseline without DMPs, and they confirm empirically on a RISC-V research core that all tested leakage classes are blocked. Evidence comes from a prototype implementation and benchmark suite, though secret-partitioning overhead can reach up to 4.8% in some cases.

What they found 1

What undercuts it the paper’s own results

Stated limits 3

What’s new here 4

Threat model stated by the source

Adversaryattacker who can monitor all microarchitectural side channels (cache, branch predictor, prefetcher, memory bus, etc.) from an intra-process sandbox, a sibling SMT context, or another core
Capabilityobserves microarchitectural state to infer secret-dependent behavior

Artifacts & methods 5

Topics

data memory-dependent prefetchingmicroarchitectural side-channelsecret taggingspeculative executionhardware prefetcherformal security proofperformance overheadarea overhead
← Previous Many AI chatbots leak conversation titles and prompts to third-party trackers