Re: [PATCH 4/6] KVM: x86: Fix determining flat mode in recalculate_apic_map

["Radim Krčmář" <rkrcmar@redhat.com>]

2014-10-01 22:16+0300, Nadav Amit:
> On Oct 1, 2014, at 9:27 PM, Radim Krčmář <rkrcmar@redhat.com> wrote:
> > Our assumption that all have the same mode is horrible.
> > (Do they all have be the same?)
> Yes: "All processors that have their APIC software enabled (using the spurious vector enable/disable bit) must have their DFRs (Destination Format Registers) programmed identically."

Thanks.

> > The only thing we allow out of your scenario that I can see is software
> > disabled x2apic after enabled clustered xapic processors and that
> > doesn't need two loops, just a sw check at x2apic.
> > Practically, it is a harmless bug :)
> So does xsa-108... ;-)
> Now seriously: First, the bug may affect certain cases of cpu hot-plug, etc.
> Second, there are additional implications. Consider a situation in which the first VCPUs have lapic disabled, and they do not have the same DFR/x2apic mode as the rest of the VCPUs.

(I coudn't find anything that would affect the host.)

> This is ok according to the SDM, but in such case, the logical map they would have would not match the spic-mode. Therefore, they may not receive NMIs, INIT, etc. - which they should regardless to the fact their LAPIC is disabled.

The guest wouldn't work, more suprising is that it usually does ;)
I'll rewrite it later if you are pressed for other stuff.