Tuning Real-Time Linux for ARM SoCs
summary
Texas Instruments tunes Real-Time Linux on ARM SoCs by testing full stacks (bootloaders, firmwares, kernels) with tools like stress‑ng, hackbench, cyclictest, perf and hardware measurements using GPIO + persistence capture. A surprising 1‑in‑a‑million RNG spin was found to cause rare latency spikes. Shrinking kernel/configuration reduces latency by roughly two‑thirds and hits hundreds of microseconds for an EtherCAT/GPIO use case, but tests run for hours, scaling across kernels/distros is manual, and CI plus reference images are needed to reproduce results reliably.
AI-written summary · about
chapters (23)
- 0:00Introduction
- 1:01Real-time Linux requirements
- 2:06Latency and determinism
- 3:31Debugging latency issues
- 5:16Automated stress testing
- 7:27Hardware vs software priorities
- 8:32Kernel size optimization
- 10:36Key optimizations summary
- 10:56Q&A
- 12:22Standardizing real-time tests
- 13:49Kernel growth challenges
- 14:32Reproducible testing
- 15:57Reference builds and CI
- 17:02Baseline measurements
- 18:05Real-time networking
- 19:36GPIO latency
- 20:39Minimizing latency deviation
- 21:34Historical context
- 22:18Real-time on laptops
- 22:59Stress test corner cases
- 24:04Debugging firmware/drivers
- 24:49Automated firmware testing
- 25:10CI and manual workflows
related talks
26mTuning Real-Time Linux for ARM SoCs
50mLow-Latency Real-Time (RT) Kernel: The Cornerstone of Arm's 5G ORAN Strategy
40mTips, Tricks, and Gotchas for Linux Real-Time Tuning
52mNot Really, but Kind of Real Time Linux
19mrtesbench: A Multi-core Benchmark Framework for Real-Time Embedded Systems
33mTowards PREEMPT_RT for the Full Task Isolation
34mPREEMPT_RT isn't Just for Lasers: The Perfect Match
28mLinux on the Arm architecture and beyond