On Wed, 2018-09-05 at 09:36 +0000, Paul Durrant wrote:
> 
> I see. Given that Windows has used APIC assist to circumvent its EOI
> then I wonder whether we can get away with essentially doing the
> same. I.e. for a completed APIC assist found in
> vlapic_has_pending_irq() we simply clear the APIC assist and highest
> vector in the ISR, rather than calling through to vlapic_EOI_set()
> and suffering the overhead. I'll spin up a patch and give it a whirl.

I think that's fine if you don't actually pass unmaskable MSIs through
to the guest in question, but if you *do* then you still need the EOI
to happen properly to "unmask" it.

Hence my approach which is basically doing what you said and bypassing
the expensive part of hvm_dpci_msi_eoi(), but *only* if there's no
unmaskable MSI to worry about.