Re: [PATCH v2 11/13] KVM: x86: add KVM_CAP_X2APIC_API

[Paolo Bonzini]

On 11/07/2016 10:56, Yang Zhang wrote:
> On 2016/7/11 15:44, Paolo Bonzini wrote:
>>
>>
>> On 11/07/2016 08:06, Yang Zhang wrote:
>>>> Changes to MSI addresses follow the format used by interrupt remapping
>>>> unit.
>>>> The upper address word, that used to be 0, contains upper 24 bits of
>>>> the LAPIC
>>>> address in its upper 24 bits.  Lower 8 bits are reserved as 0.
>>>> Using the upper address word is not backward-compatible either as we
>>>> didn't
>>>> check that userspace zeroed the word.  Reserved bits are still not
>>>> explicitly
>>>
>>> Does this means we cannot migrate the VM from KVM_CAP_X2APIC_API enabled
>>> host to the disable host even VM doesn't have more than 255 VCPUs?
>>
>> Yes, but that's why KVM_CAP_X2APIC_API is enabled manually.  The idea is
>> that QEMU will not use KVM_CAP_X2APIC_API except on the newest machine
>> type.
> 
> Thanks for confirmation. And when the KVM_CAP_X2APIC_API will be enabled
> in Qemu?

It could be 2.7 or 2.8.

>>
>> If interrupt remapping is on, KVM_CAP_X2APIC_API is needed even with 8
>> VCPUs, I think.  Otherwise KVM will believe that 0xff is "broadcast"
>> rather than "cluster 0, CPUs 0-7".
> 
> If interrupt remapping is using, what 0xff means is relying on which
> mode the destination CPU is in. I think there is no KVM_CAP_X2APIC_API
> needed since interrupt remapping table gives all the information.

If you have EIM 0xff never means broadcast, but KVM sees a 0xff in the
interrupt route or KVM_SIGNAL_MSI argument and translates it into a
broadcast.

Paolo