Do Existing Drivers Conflict with PREEMPT_RT? a 3-Year Survey
summary
A three year survey catalogues non-core PREEMPT_RT driver diffs and shows concrete fixes: sequence counters gain extra runtime sanity checks and a latch variant to avoid read-side infinite loops; tasklets trigger a manual audit of 400+ drivers where migration from busy-wait tasklet_kill to condition variables and sleeping removes a kernel deadlock; tooling uses coccinelle for simple patterns but requires manual review for dynamic context checks; results report only 10 drivers already safe and recommend using standard kernel locking APIs, not open-coded context checks.
AI-written summary · about
chapters (9)
- 0:00Talk introduction
- 1:31PREEMPT_RT benefits
- 4:01Sequence counters challenges
- 10:08Tasklets API issues
- 15:37Tasklets removal and compliance
- 18:52PREEMPT_RT with standard APIs
- 23:29Printk and console locking
- 25:16Q&A
- 29:14PREEMPT_RT vs static partitioning
related talks
34mDo Existing Drivers Conflict with PREEMPT_RT? a 3-Year Survey
53mPREEMPT_RT - How Not to Break It.
24mPREEMPT_RT Q&A with tglx
47mWhat Every Driver Developer Should Know about RT
50mInside the RT Patch
26mReal-time Linux: what is next?
40mRT Linux in Safety-Critical Systems: The Potential and the Challenges
39mEvaluation of PREEMPT_RT in Virtualized Environments