BootRIST: Detecting and Isolating Mercurial Cores at the Booting Stage

With the shrinking of transistor size and their own increasing complexity, CPUs have become more fault-prone. The mercurial cores, one type of processor failure, can randomly return silent miscomputation results and have been discovered in commercial CPUs. To address the mercurial core problem, we propose BootRIST, a software-based solution to detect and isolate mercurial cores in CPUs at the booting stage. BootRIST provides a framework to execute the testing instructions and store the result for each CPU core. Based on the execution result, BootRIST leverages the CPU hotplug mechanism to isolate mercurial cores. With BootRIST, the system and software can correctly run on CPUs with mercurial cores. We have implemented a prototype of BootRIST and deployed it on an Arm Linux server to perform an extensive evaluation. The empirical studies demonstrate that BootRIST can effectively detect and isolate mercurial cores at the booting stage.