KVM Live Upgrade with Properly Handling of Passthrough Devices
summary
Implements a KVM live-upgrade that preserves passthrough devices by loading parallel kernel modules (k1/k2), inheriting VF connectors, and keeping GPA→HPA mappings so device DMA continues; uses posted-interrupt descriptor sharing to avoid lost IRQs and precreates device structures plus concurrent save/load with seed memory to shrink downtime. Measured vCPU stop time ranges 11–38 ms across stress, memtester and fio workloads, showing the method works with modest pause times for cloud VMs.
AI-written summary · about
chapters (5)
related talks
14mDeep Optimization of VMM Live Upgrade
33mPreserving IOMMU States During Kexec Reboot
24mSupport for Fast and Reliable VMM Live Upgrades in Libvirt
22mQEMU/KVM Upgrade Test (Stable Guest ABI and in Place Upgrade)
10mTowards a More Efficiently Synchronization in KVM
33mLinux Kernel Live Patching
28mHyperscale Kernel Live Patching at Meta
39mKVM Modifications for Clustered Operation