Linux 5.18's KVM Squeezes In AMD Nested Virtualization Improvements

Last week saw the main KVM virtualization pull for the Linux 5.18 feature updates while sent in today was a second batch of improvements for the Kernel-based Virtual Machine.

First up, Xen event channel acceleration is added for KVM. This patch series goes back a while and allows guest vCPUs to send each other IPIs without having to go through the user-space VMM.

There are a number of AMD nested virtualization improvements for this secondary pull request. Among the AMD nested virtualization improvements are optimizing "nested nested" optimizations (nesting a VM within an already nested virtual machine) when it comes to the VMLOAD/VMSAVE and nested vGIF handling. There is also support for allowing AVIC to co-exist with a nested guest running, fixes for LBR virtualization, PAUSE filtering for nested hypervisors, and other improvements around nested virtualization on AMD Ryzen/EPYC platforms.

Some of the other KVM updates for today's pull request includes MSR register filtering is now only done for MSRs being accessed via the RDMSR/WRMSR instructions, support for user-space to opt-out of hypercall patching, and new kernel interfaces to get/set the TSC frequency for an entire virtual machine. There are also fixes around PMU virtualization, preventing the KVM module from exiting until all VMs are freed, and other general bug fixes.

More details on this batch of KVM virtualization updates for Linux 5.18 via this PR from KVM maintainer Paolo Bonzini of Red Hat.