Reliably reproducing kernel data races
summary
Introduces the Linux Test Project Fuzzy Sync C library, which reproduces kernel data races from user space by colliding events without special kernel instrumentation. Demonstrates a contrived user-space race, then shows reproducing real kernel races including using a delay bias to force event ordering, and explains using Fuzzy Sync for regression tests across kernels and other contexts.
AI-written summary · about
related talks
18mAccurate Data Race Prediction in the Linux Kernel through Sparse Fourier Learning
28mExtreme OS Kernel Testing
44m"go test -race" Under the Hood
22mData race detection
46mBypass Kernel Barriers: Fuzzing Linux Kernel in Userspace With LKL
15mSound and partially-complete static analysis of data-races in GPU programs
52mFantastic C++ Bugs and How to Test Them
40mRT Linux in Safety-Critical Systems: The Potential and the Challenges