All Bark and no Bite: vCPU Stall Detection for KVM Guests
summary
A para‑virtual vCPU stall detector implements a per‑vCPU MMIO watchdog: crosvm exports per‑CPU registers, the guest writes periodic heartbeats to a Linux kernel front-end driver, and an internal per‑vCPU counter decrements until zero to declare a stall. The design reads /proc/stat to account stolen time, handles CPU hotplug by disarming HR timers, and is upstream with crosvm patches merged and the kernel driver accepted after 12 revisions; planned Android integration and VM diagnostics reporting are next.
AI-written summary · about
chapters (4)
related talks
45mDynamic vCPU priority boosting in KVM for latency sensitive workloads
37mHardware Error Handling Improvement for Reliable KVM Hypervisor
25mChallenges in Supporting Virtual CPU Hotplug on SoC Based Systems (like ARM64)
23mHyper-V's Virtual Secure Mode in KVM project update
28mCPUfreq/sched and VM guest workload problems
31mKVM Nested Virtualization: Workings and Usecases
09mKVM Latency Performance Tuning
39mEvaluation of PREEMPT_RT in Virtualized Environments