On 2/25/25 00:55, Sean Christopherson wrote:
This was _supposed_ to be a tiny one-off patch to fix a nVMX bug where KVM
fails to detect that, after nested VM-Exit, L1 has a pending IRQ (or NMI).
But because x86's nested teardown flows are garbage (KVM simply forces a
nested VM-Exit to put the vCPU back into L1), that simple fix snowballed.
The immediate issue is that checking for a pending interrupt accesses the
legacy PIC, and x86's kvm_arch_destroy_vm() currently frees the PIC before
destroying vCPUs, i.e. checking for IRQs during the forced nested VM-Exit
results in a NULL pointer deref (or use-after-free if KVM didn't nullify
the PIC pointer). That's patch 1.
Patch 2 is the original nVMX fix.
The remaining patches attempt to bring a bit of sanity to x86's VM
teardown code, which has accumulated a lot of cruft over the years. E.g.
KVM currently unloads each vCPU's MMUs in a separate operation from
destroying vCPUs, all because when guest SMP support was added, KVM had a
kludgy MMU teardown flow that broken when a VM had more than one 1 vCPU.
And that oddity lived on, for 18 years...
Queued patches 1 and 2 to kvm/master, and everything to kvm/queue
(pending a little more testing and the related TDX change).
Paolo
