← Lab /Software

DualGuard cuts memory-safety overhead up to 72% while protecting speculative and non-speculative code

Software2026-10-29T05:04:48Z Method · Prototype · arXiv Threat model

The claim

DualGuard provides dual memory safety with lower performance overhead than naive combinations of existing defenses

Sixuan Dang, Quan Zhou, Kaiming Huang, Jianan Zhu, Anitha Gollamudi, Trent Jaeger, Danfeng Zhang · arXiv Paper PDF

Published as: “DualGuard: Comprehensive and Efficient Memory Safety across Sequential and Speculative Execution”

So what

Deploying DUALGUARD can harden existing software against both classic out-of-bounds bugs and Spectre-style speculative attacks with modest performance cost.

The five-minute version

DUALGUARD achieves dual memory safety with substantially lower performance overhead than naïve stacking of sequential and speculative defenses, cutting average overhead from 148.4% to 43.6% for Baggy, from 119.7% to 33.6% for Delta Pointers, and from 103.2% to 99.5% for SafeStack. The paper provides a formal safety-preservation framework and implements dual-safe variants (DUALBAGGY, DUALDELTAPOINTERS) that eliminate redundant checks. Evidence comes from SPEC CPU 2017 and real-world cryptographic library benchmarks, though some configurations see only modest gains and a few cases incur slightly higher overhead.

What they found 2

Baggy+SLH overhead reduced from 148.4% to 43.6% with DUALGUARD-DUALBAGGY
“On average,3 BAGGY+SLH incurs an overhead of 148.4%, whereas DUALGUARD-DUALBAGGY reduces it to 43.6%.”
DeltaPointers+SLH overhead reduced from 119.7% to 33.6% with DUALGUARD-DUALDELTAPOINTERS
“On average, DELTAPOINTERS+SLH incurs an overhead of 119.7%, whereas DUALGUARD-DUALDELTAPOINTERS reduces it to 33.6%.”

What undercuts it the paper’s own results

DUALGUARD-DUALDELTAPOINTERS consistently incurs lower overhead except for 2 benchmarks where it incurs a slightly higher overhead.
The reduction for SafeStack-based protection is smaller than what we observe in the Baggy-based and Delta-Pointers-based cases.

Stated limits 3

We leave improving the scalability of our prototype as future work.
Temporal Memory Safety: Temporal defenses ... Under speculative execution, stale or mispredicted metadata could lead to transient use-after-free conditions or premature object reclamation.
Other speculative attacks: ... such indirect mispredictions could speculatively bypass memory-safety checks entirely ... These attack surfaces introduce failure modes that are qualitatively distinct from the bounds-check bypass scenarios addressed in this paper.

What’s new here 4

Formal safety-preservation property enabling sound composition of sequential and speculative memory-safety mechanisms.
Dual-safe variants of existing defenses (DUALBAGGY, DUALDELTAPOINTERS) that are provably safe under both execution models.
Selective composition strategy that omits unnecessary checks based on violation-region analysis, reducing overhead.
Comprehensive evaluation showing up to 95% overhead reduction compared to naïve compositions.

Threat model stated by the source

Adversaryattacker who can fully exploit speculative program execution, including controlling the branch predictor and providing malicious inputs

Capabilitycan induce speculative out-of-bounds memory accesses and exploit any resulting transient accesses via cache side channels, Rowhammer, or ROP

Assumesfocuses on spatial memory-safety violations only; targets Spectre v1 (bounds-check bypass) attacks; temporal errors are out of scope

Artifacts & methods 7

DUALGUARDsystem
DUALBAGGYprimitive
DUALDELTAPOINTERSprimitive
static program analysis for violation-region identification
formal safety-preservation frameworktheory
SPEC CPU 2017 benchmark suitedataset
cryptographic library benchmark setdataset

Topics

spatial memory safetyspeculative executionSpectre v1safety preservationdual memory safetystatic analysisLLVMselective composition

← Previous ARBITRA achieves higher precision in document-level ATT&CK extraction and halves precision loss under…