Stephen McCamant

Kartik Ramkrishnan, Stephen McCamant, Antonia Zhai et al.

Cache partitioning is among the strongest structural defenses against eviction-based cache side channels, yet a decade-old design issue has blocked its widespread deployment in secure shared-OS settings. The issue is that write-shared coherence collapses under strict partitioning. We present SCP (Secure and Coherent Partitioning), which combines strict eviction isolation with write-shared coherence by partitioning only the tags, sharing a single data pool, and sizing the data pool so capacity-driven cross-partition eviction cannot occur. Timing obfuscation extends protections to the inter-partition lookup path. Coherence-based leakage on shared-writeable lines is mitigated by routing those writes through to the LLC once a leakage threshold is crossed, which makes attacker write probe latency independent of victim activity. Using gem5 for implementation, SCP mitigates Prime+Probe and Flush+Reload, which are the basis for more sophisticated cache attacks. We also demonstrate that a shared-writeable-line attack is mitigated. All these attacks yield results no better than random guessing. SCP's hardware cost is a modest +2.8% LLC SRAM. Performance matches DAWG within 0.3% IPC on the SPEC CPU2017 benchmarks that we evaluated. Sharing-intensive microbenchmarks demonstrate a tunable security-performance tradeoff based on a system-specified leakage threshold.

Andrew Bao, Haochen Zeng, Peng Chen et al.

One-day vulnerabilities pose significant risks due to delayed or incomplete patch adoption. Generating proof-of-concept (PoC) inputs is therefore essential for assessing real-world impact. The key challenge is identifying necessary constraints for triggering the vulnerability and solving them effectively. Existing directed fuzzing approaches prioritize inputs toward target locations, but neither explicitly identify necessary constraints nor solve them effectively, relying instead on target-distance feedback and random mutation. Agentic approaches show strong potential through code reasoning and structured input generation, but goal drift in long-horizon reasoning limits their effectiveness. DIG addresses this challenge by exploiting a key property of one-day vulnerabilities: patches often reveal necessary preconditions for triggering. DIG uses an LLM to analyze the patch and synthesize an oracle making these conditions explicit. The oracle supports effective PoC generation at two levels. At the high level, DIG performs oracle-guided generator evolution, where an agent infers and solves constraints to satisfy the oracle. At the low level, DIG instruments the oracle into the target program and uses branch-distance feedback to guide random mutation in directed fuzzing. Evaluation shows DIG outperforms 2 state-of-the-art agents and 10 fuzzers across 138 real-world CVEs. DIG triggers 80 vulnerabilities, surpassing prior results and outperforming the best baseline by 40% (57 vs. 80 CVEs). Notably, DIG exclusively triggers 9 vulnerabilities no existing technique can trigger. Compared to the average of other tools, DIG triggers vulnerabilities faster in 92.9% of cases, achieving over 100x speedup in 48.8% of cases, with a maximum speedup of 3,664x. Beyond one-day PoC generation, DIG uncovers 6 previously unknown vulnerabilities in widely deployed libraries, enabling zero-day discovery.

Kartik Ramkrishnan, Antonia Zhai, Stephen McCamant et al.

The last level cache is vulnerable to timing based side channel attacks because it is shared by the attacker and the victim processes even if they are located on different cores. These timing attacks evict the victim cache lines using small conflict groups(SCG), and monitor the cache to observe when the victim uses these cache lines again. A conflict group is a collection of cache lines which will evict the target cache line. Randomization is often used by defenses to prevent creation of SCGs. We introduce new attacks to demonstrate that the current randomization schemes require an extremely high refresh rate to be secure, on average a 15\% performance overhead, and upto 50\% in the worst case. Next, we propose a new randomization strategy using an indirection table, which mitigates this issue. Addresses of cache lines are encrypted and used to lookup the indirection table entry. Each indirection table entry stores a mapping to a randomly chosen cache set. The cache line is placed into this randomly chosen set. The encryption key changes upto 50x faster than CEASER's default rate, by using evictions to trigger the re-randomization. Instead of moving cache lines, this mechanism re-randomizes one iTable entry at a time, whenever the cache lines corresponding to the iTable entry are naturally evicted. Thus, the miss rate is not much worse than the baseline. We quantitatively show that our scheme does almost as well as a fully associative cache to defend against these attacks. We also demonstrate new attacks that target the iTable by oversubscribing its entries, and quantitatively show that our scheme is resilient against new attacks for trillions of years. We estimate low area ( < 7\%) and power overhead compared to a baseline inclusive last-level cache. Lastly, we evaluate a low performance overhead (<4%) using the SPECrate 2017 and PARSEC 3.0 benchmarks.

Seonmo Kim, Stephen McCamant

Approximate model counting for bit-vector SMT formulas (generalizing \#SAT) has many applications such as probabilistic inference and quantitative information-flow security, but it is computationally difficult. Adding random parity constraints (XOR streamlining) and then checking satisfiability is an effective approximation technique, but it requires a prior hypothesis about the model count to produce useful results. We propose an approach inspired by statistical estimation to continually refine a probabilistic estimate of the model count for a formula, so that each XOR-streamlined query yields as much information as possible. We implement this approach, with an approximate probability model, as a wrapper around an off-the-shelf SMT solver or SAT solver. Experimental results show that the implementation is faster than the most similar previous approaches which used simpler refinement strategies. The technique also lets us model count formulas over floating-point constraints, which we demonstrate with an application to a vulnerability in differential privacy mechanisms.

Vaibhav Sharma, Kesha Hietala, Stephen McCamant

Independently developed codebases typically contain many segments of code that perform same or closely related operations (semantic clones). Finding functionally equivalent segments enables applications like replacing a segment by a more efficient or more secure alternative. Such related segments often have different interfaces, so some glue code (an adapter) is needed to replace one with the other. We present an algorithm that searches for replaceable code segments at the function level by attempting to synthesize an adapter between them from some family of adapters; it terminates if it finds no possible adapter. We implement our technique using (1) concrete adapter enumeration based on Intel's Pin framework (2) binary symbolic execution, and explore the relation between size of adapter search space and total search time. We present examples of applying adapter synthesis for improving security and efficiency of binary functions, deobfuscating binary functions, and switching between binary implementations of RC4. We present two large-scale evaluations, (1) we run adapter synthesis on more than 13,000 function pairs from the Linux C library, (2) using more than 61,000 fragments of binary code extracted from a ARM image built for the iPod Nano 2g device and known functions from the VLC media player, we evaluate our adapter synthesis implementation on more than a million synthesis tasks . Our results confirm that several instances of adaptably equivalent binary functions exist in real-world code, and suggest that adapter synthesis can be applied for reverse engineering and for constructing cleaner, less buggy, more efficient programs.