Proposing a new tracer to monitor RT task behavior
summary
Proposal builds a practical tracer implemented as a BPFTrace script (~100 lines) that monitors RT tasks and flags specific mistakes: it reports page faults and S-state sleeps, notes when tasks are preempted 390 µs (demo KVM numbers), and detects non-PI futexes and timerfd usage. The tool prints live events (page faults, unintended sleeps, preemption durations), runs in-kernel via kprobes to gather real RT priority info, and the demo works end-to-end, showing cyclictest interrupts and revealing setup-time page faults.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 2:28Preempt RT mistakes
- 11:14Real-time Linux challenges
- 16:46Built-in RT tracer proposal
- 21:48Live BPF Trace demo
- 32:30Demo results analysis
- 41:01Tracer impact on development
- 44:55Future enhancements
- 51:36Q&A
related talks
21mRTLA Requests and TODOs
31mFantastic tracepoints and where to find them
09mrtla: what is next?
34mTracing: The Bane of You Security Folks
40mRT Linux in Safety-Critical Systems: The Potential and the Challenges
35mLinux Tracing updates
18mScheduler Tracing With ftrace + eBPF
50mrtla timerlat: Debugging Real-Time Linux Scheduling Latency