On Sat, 05 Jun 2021 11:03:40 +0100,
Sergey Senozhatsky <senozhatsky@xxxxxxxxxxxx> wrote:
>
> On (21/06/05 18:58), Sergey Senozhatsky wrote:
> [..]
> > > > +static int kvm_arch_suspend_notifier(struct kvm *kvm)
> > > > +{
> > > > + struct kvm_vcpu *vcpu;
> > > > + int i, ret;
> > > > +
> > > > + mutex_lock(&kvm->lock);
> > > > + kvm_for_each_vcpu(i, vcpu, kvm) {
> > > > + ret = kvm_set_guest_paused(vcpu);
> > > > + if (ret) {
> > > > + pr_err("Failed to pause guest VCPU%d: %d\n",
> > > > + vcpu->vcpu_id, ret);
> > >
> > > Is it really a good idea to fail suspend when a guest doesn't have PV
> > > time enabled? I also wonder how useful the pr_err() is, given that it
> > > contains no information that would help identifying which guest failed
> > > to pause.
> >
> > No opinion. What shall we do when we fail to suspend the VM?
> > VM's watchdogs will trigger and maybe panic the system after
> > resume.
Panic the guest, maybe. Panic the host, surely not? You cannot decide
what the guest uses or doesn't use anyway, so that's out of your
hands. I don't think this should prevent the host from suspending, in
any case.
> For the time being kvm_set_guest_paused() errors out when
> !vcpu->arch.pv_time_enabled, but this probably can change in the
> future (who knows?). So shall I check vcpu->arch.pv_time_enabled in
> kvm_arch_suspend_notifier()?
That, or check for the -EINVAL return value.
M.
--
Without deviation from the norm, progress is not possible.
