Speaker / Predavač
Naslov / Title
Hardware-assisted data race detection with COEMS
Sažetak / Abstract
Detecting data races is a challenge in modern code that harnesses multi-core CPUs. Instrumentation-based techniques have a high performance impact, and are not likely to be certified for safety-critical systems. The COEMS project follows a hardware-assisted approach, where hardware tracing facilities present on modern processors are used, combined with a separate hardware component to monitor the trace. We describe a data race detector based on the well-known lock-set algorithm in the runtime verification language TeSSLa. The TeSSLa pattern for race detection is instantiated for a particular number of threads, locks, and memory locations, and can then be synthezised to the FPGA-based hardware observer platform. We develop static analysis tool that can be used to obtain the number of threads and locks.
date/vreme: Tuesday, 06/03/2018, 11:00-12:00 h
venue/mesto: sala za sastanke na 3. spratu u KULI FTN