Re: [PATCH 4/4][RFC] kvm: eoi_eventfd

[Alex Williamson <alex.williamson@redhat.com>]

On Thu, 2012-06-28 at 19:27 +0300, Avi Kivity wrote:
> On 06/24/2012 06:02 PM, Alex Williamson wrote:
> > On Sun, 2012-06-24 at 15:56 +0300, Avi Kivity wrote:
> >> On 06/23/2012 01:16 AM, Alex Williamson wrote:
> >> > I think we're probably also going to need something like this.
> >> > When running in non-accelerated qemu, we're going to have to
> >> > create some kind of EOI notifier for drivers.  VFIO can make
> >> > additional improvements when running on KVM so it will probably
> >> > make use of the KVM_IRQFD_LEVEL_EOI interface, but we don't
> >> > want to have a generic EOI notifier in qemu that just stops
> >> > working when kvm-ioapic is enabled.
> >> 
> >> Why?
> > 
> > Hmm, I must be missing something or not describing it correctly, because
> > it seems obvious.  
> 
> I have not exhausted this quarter's quota of stupid questions yet.

;)

> > If we create a dependency in qemu of needing to know
> > when an eoi occurs and notifier a driver and have no way to fulfill that
> > dependency when running on kvm... that'd be bad, right?  I don't want to
> > assume that every consumer of such an interface would prefer to make use
> > of an irqfd.  Not sure if that answers your question though.  Thanks,
> 
> I meant, what scenario do you have in mind where we want the EOI
> notifier while running with kvm-irqchip enabled?  Perhaps I phrased my
> question a bit too tersely.

Aha.  Well, the v2 series really pulls eoifd in as more of a primary
player and is now the only way to get EOI notification regardless of
whether it's bound to a separate level irq source or used by qemu for
the common userspace source.  I don't know of any users that currently
need EOI notification for the common userspace source id, but depending
on where we land with whether KVM_IRQFD supports level interrupts, it
may be vfio that needs it.

Alex