Kernel handling of CPU and memory hot un/plug events for crash
summary
Kernel now registers a generic crash notifier on CPU and memory notifiers and updates only the affected elfcorehdr segments in-kernel instead of triggering full kdump reloads via udev; kexec-tools gets a new flag and elfcorehdr is excluded from SHA so the kernel can safely patch it. Measured downtime drops from ~1s to ~4ms for CPU hotplug and ~0.03ms for memory; a sysfs bit advertises kernel-side handling and the feature is upstream with a CONFIG option.
AI-written summary · about
chapters (10)
- 0:00Introduction and motivation
- 1:08Kdump overview
- 6:25Impact of hot plug on Kdump
- 7:53Existing user-space solution
- 11:04Proposed kernel-based solution
- 13:33Implementation details
- 18:08Kernel-side changes
- 19:10User-space tool updates
- 20:12Enabling in other architectures
- 22:45Q&A session
related talks
37mHardware Error Handling Improvement for Reliable KVM Hypervisor
52mLinux Kernel Debugging: Going Beyond Printk Messages
43mEfficient and Practical Capturing of Crash Data on Embedded Systems
13mRamdump for Trusted VMs: Efficient Crash Debugging Without Reboot Device
26mStateless Linux Kernel Crashdumps
19mKdump memory usage
37mBreaking and Fixing the Kernel: Crashes, Debugging & Performance Tun...
49mKernel dump analysis made quick and easy!