On Wed, 14 Jun 2017 10:22:16 -0300 Eduardo Habkost <ehabk...@redhat.com> wrote:
> On Wed, Jun 14, 2017 at 03:17:54PM +0200, Paolo Bonzini wrote: > > > > > > On 14/06/2017 15:11, Igor Mammedov wrote: > > >> No, KVM really uses the VCPU _index_ for HV_X64_MSR_VP_INDEX: > > > > > > and as you pointed out that works just by luck, > > > as soon as we there would be out of order created CPUs > > > returned value won't match cpu_index. > > > > > > So instead of spreading this nonsense out to QEMU, is it possible > > > to fix kernel(kvm+guest) to use apic_id instead? > > > > Yes, definitely. At this point let's add a new "KVM_CAP_HYPERV_SYNIC2" > > capability and declare the old one broken, that will make things easier. > > What do we need a capability for? > Can't we just call > KVM_SET_MSRS using arch_id (== apic_id) and warn the user if it > doesn't work? warn won't solve issue, and it's called rather late in vcpu creation chain without any error propagation so it's not possible (hard) to gracefully fail cpu hotplug. with KVM_CAP_HYPERV_SYNIC2 we could easily turn on compat property for SYNC device (needs QOMification) that would turn on broken compat logic if capability is not present/old machine type.
