Hardware Error Handling Improvement for Reliable KVM Hypervisor
summary
Three concrete fixes and open patches aimed at making KVM survive CPU/memory machine checks: adds a makedumpfile exclusion for hardware-poison pages so vmcore creation no longer touches poisoned memory (demo excludes ~20 poisoned pages), implements logging/grind logic in mcelog/MC log to track corrected-error hotspots, and proposes changing the poison handler to avoid kernel panics that cause data loss of dirty cache. Patches are upstream-discussed and partially merged; guest I/O cache truncation still requires policy work.
AI-written summary · about
chapters (9)
- 0:00Introduction and background
- 2:20Hardware error handling scope
- 6:12Expectations and challenges
- 10:06Error isolation mechanisms
- 15:13Isolation process breakdown
- 19:32Addressing dump and data loss
- 23:30Logical data loss analysis
- 29:11Error logging and future needs
- 34:24Conclusion and future work
related talks
26mLessons Learned: Optimizing KVM Performance for EHR Systems
27mDay 2 Keynote Sessions
31mKernel handling of CPU and memory hot un/plug events for crash
40mCurrent Status and Future of HWPOISON Subsystem
50mPanel Discussion: KVM-based Virtualization Contributor Q&A
16mUnifying KVM API for protected VM and utilities
38mKernel Hardening: Protecting the Protection Mechanisms
42mProtecting the Linux Kernel from Confidential Computing Threats in Virtual Environments