Tips, Tricks, and Gotchas for Linux Real-Time Tuning
Talkat Embedded Linux Conference @ Open Source Summit North America 2022 · on Wednesday, 2022-06-22· at 16:35· Room 201/202 (Level 2), JW Marriott Austin, 110 E 2nd Street Austin, TX 78701·
summary
Practical checklist: run cyclictest (with realistic IO load) and plot histograms; collect traces with trace-cmd and inspect in kernelshark; use perf/BPF for hotspots. Install PREEMPT_RT, partition or isolate CPUs, mlockall, disable CPU freq and deep C‑states, and consider turning off RT throttling. Watch for SMI-induced spikes and MMIO CPU stalls (seen as ~400µs tails); fix driver IRQs with request_threaded_irq and avoid priority inversions via priority inheritance.
AI-written summary · about
chapters (11)
- 0:00Start
- 3:20Measuring Latency What Kind of Tools Do You Use
- 10:55Partitioning Your Cpus
- 12:16Set the Scheduling Priority
- 14:46Power Management
- 15:59Memory Corrections
- 16:45Safety Nets
- 18:14Safety Nets for Memory
- 18:44Prioritize Processes
- 19:52Gotchas
- 32:42Summary
related talks
26mTuning Real-Time Linux for ARM SoCs- 26mTuning Real-Time Linux for ARM SoCs
47mWhat Every Driver Developer Should Know about RT
39mHow to Investigate Issues on Linux Embedded System?
50mLow-Latency Real-Time (RT) Kernel: The Cornerstone of Arm's 5G ORAN Strategy
52mNot Really, but Kind of Real Time Linux
51mAdventures in (Simulated) Asymmetric Scheduling
46mTools and Techniques to Debug an Embedded Linux System