PhantomCache: Obfuscating Cache Conflicts with Localized Randomization

Abstract

Cache conflicts due to deterministic memory-tocache mapping have long been exploited to leak sensitive information such as secret keys. While randomized mapping is fully investigated for L1 caches, it still remains unresolved about how to secure a much larger last-level cache (LLC). Recent solutions periodically change the mapping strategy to disrupt the crafting of conflicted addresses, which is a critical attack procedure to exploit cache conflicts. Remapping, however, increases both miss rate and access latency. We present PhantomCache for securing an LLC with remapping-free randomized mapping. We propose a localized randomization technique to bound randomized mapping of a memory address within only a limited number of cache sets. The small randomization space offers fast set search over an LLC in a memory access. The intrinsic randomness still suffices to obfuscate conflicts and disrupt efficient exploitation of conflicted addresses. We evaluate PhantomCache against an attacker exploring the state-of-the-art attack with linear-complexity. To secure an 8-bank 16 MB 16-way LLC, PhantomCache confines randomization space of an address within 8 sets and brings only 1.20% performance degradation on individual benchmarks, 0.50% performance degradation on mixed workloads, and 0.50% storage overhead per cache line, which are 2x and 9x more efficient than the state-of-the-art solutions. Moreover, PhantomCache is solely an architectural solution and requires no software change. * Qinhan Tan and Zhihua Zeng have contributed equally and are considered to be co-first authors.

Systematic Analysis of Randomization-based Protected Cache Architectures

“…Our work provides a baseline to compare future secure caches and their analysis. Future work should investigate how our techniques can be applied to concurrent work [46]. This paper also shows the importance of cryptanalytic resistance of the core randomization mapping.…”

ClepsydraCache -- Preventing Cache Attacks with Time-Based Evictions

“…One of the most recent approaches PhantomCache [27] is a pure architectural solution and uses a localized randomization technique to bound the randomized mapping to a limited number of cache sets. The randomization technique used in PhantomCache allows an address to be mapped to multiple locations within a single cache bank.…”

“…However, the security goals we pursue with our design define important aspects of the concept and hence, we now define the attacker model. Our threat model is similar to previous work in this field [26], [27]: we consider a strong adversary model, i.e., a noise-free scenario, where the attacker is able to perfectly distinguish between a cache hit and a cache miss for any memory address. As in real-world attacks, the attacker has no insights on the internal state of the cache except those leaked by the timing differences of memory accesses.…”

“…We hold on to the traditional set-associative design and introduce a variant of the skewing approach which was originally proposed in 1993 [41]. As studied in recent years, randomized skewing makes finding conflicting addresses much more challenging for an attacker, see [26], [27]. In combination with the time-based eviction, finding eviction sets becomes even harder as motivated in Section VI.…”

“…We show, that CLEPSYDRACACHE offers strong security properties and minimizes the information an attacker can learn about conflicting addresses. In contrast to other concepts that solely focus on index randomization (see [26], [27]) CLEPSYDRACACHE hides cache conflicts from an attacker and due to the randomness of the TTL, attacks measuring the execution time of a program experience large noise distortions. We also analyze the security of CLEPSY-DRACACHE against attackers using a minimal eviction set or the clflush instruction from the x86 instruction set.…”