On Wed, 2018-09-05 at 10:40 +0000, Paul Durrant wrote:
> 
> Actually the neatest approach would be to get information into the
> vlapic code as to whether APIC assist is suitable for the given
> vector so that the code there can selectively enable it, and then Xen
> would know it was safe to avoid fully emulating an EOI for anything
> that did have assist enabled.

I'm not sure I understand why an assisted EOI should be any different
from "normal" EOI.

The global lock and indirect function calls and all that stuff that
hvm_dpci_msi_eoi() does are *expensive*, for a rare case.

Why not bypass all that when it's not needed, for "normal" EOI and
APIC-assisted EOI alike? Why have a distinction between the two?