Look into notifying the guest when vcpus are paused

[alxiord]

According to the documentation:

KVM_KVMCLOCK_CTRL

This signals to the host kernel that the specified guest is being paused by
userspace.  The host will set a flag in the pvclock structure that is checked
from the soft lockup watchdog.  [...] This ioctl can be called any time
after pausing the vcpu, but before it is resumed.

KVM uses this ioctl to notify the guest that it's being paused(code):

case KVM_KVMCLOCK_CTRL: {
    r = kvm_set_guest_paused(vcpu);
    goto out;
}

/*
 * kvm_set_guest_paused() indicates to the guest kernel that it has been
 * stopped by the hypervisor.  This function will be called from the host only.
 * EINVAL is returned when the host attempts to set the flag for a guest that
 * does not support pv clocks.
 */
static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
{
	if (!vcpu->arch.pv_time_enabled)
		return -EINVAL;
	vcpu->arch.pvclock_set_guest_stopped_request = true;
	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
	return 0;
}

In the context of snapshotting a microVM and resuming it later, this ioctl needs to be evaluated to assess its benefits & costs and whether it should be called immediately after pausing or between deserialization and resuming.

[acatangiu]

Adding a bit more detail here:

In a case where the guest is paused by the hypervisor for a non-negligible period of time, that blackout period is seen by the guest as frozen/non-ticking CPUs. On resuming the VM, the guest soft-lockup watchdog might panic.

This ioctl makes KVM on the host set a special flag on the emulated/paravirtualized pv-clock. This flag is seen by the guest and tells the guest that the vCPUs haven't ticked in a while on purpose.

[alindima]

On resuming the VM, the guest soft-lockup watchdog might panic.

Remember seeing any such panics?

[wearyzen]

This comment in our code looks related

firecracker/src/vmm/src/vstate/vcpu/mod.rs

Line 299 in cb2ea92

// TODO: we should call `KVM_KVMCLOCK_CTRL` here to make sure