Want to find and understand side channel leakage in your software? Check out our @USENIXSecurity 23 paper “CacheQL: Quantifying and Localizing Cache Side Channel Vulnerabilities in Production Software”, an automated and scalable way to detect and quantify side channel leakage.

12/n Our tool is available at: https://t.co/81yBMCxkxy https://t.co/EKK2sRq3bd It's fully end-to-end; you only need to provide the secret and its derived side channel traces. You can try to analyze other software and other formats of side channels.

11/n We quantified the leaked information in various software under different settings.

10/n To localize leakage sites, we reformulate information leaks via a side channel trace as a cooperative game among all records on that trace. Then, by computing the Shapley value, we allocate the quantified leaks among program points, thereby localizing those leaking secrets.

9/n With conditional probability, we also principally alleviate the coverage issue of conventional dynamic methods.

8/n The information leakage can be quantified using mutual information (MI) between secrets (e.g., a private key, an input image) and the corresponding side channels. To improve the precision and computing efficiency, we recast computing MI as estimating conditional probability.

7/n We first compare existing works using the above criteria. Then, we design CacheQL which satisfies all the eight criteria. CacheQL has two main functionalities: quantification and localization.

6/n 5) localize vulnerable program points leaking secrets; 6) analyze leakage of secret keys and media data; 7) scalable for whole-program/whole-trace analysis; 8) consider both implicit and explicit information flow.

5/n 1) analyze both execution traces and real-world cache attack logs; 2) analyze both deterministic and non-deterministic observations; 3) be able to analyze program executables; 4) deliver both qualitative and quantitative analysis;

4/n Our work initializes a comprehensive view on detecting cache side-channel vulnerabilities. We propose eight criteria to design a full-fledged detector.

3/n We categorize these vulnerabilities into five categories: a) leaking secrets in pre-processing stage; b) leaking secrets in decryption stage; c) leaking leading zeros; d) leaking secrets via explicit information flow; e) leaking secrets via implicit information flow.

2/n With CacheQL, we identified hundreds of new leakage sites in OpenSSL, MbedTLS, Libgcrypt, and Libjpeg. The full report of our localized vulnerabilities can be found at: