Re: [PATCH v8 6/6] kvm: Add de-assert option to KVM_IRQ_ACKFD

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

On Wed, 2012-08-15 at 17:11 +0300, Michael S. Tsirkin wrote:
> On Fri, Aug 10, 2012 at 04:37:56PM -0600, Alex Williamson wrote:
> > It's likely (vfio) that one of the reasons to watch for an IRQ ACK
> > is to de-assert and re-enable an interrupt.  As the IRQ ACK notfier
> > is already watching a GSI for an IRQ source ID we can easily couple
> > these together.
> > 
> > Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
> 
> This source id is required the only way to assert
> in the 1st place is with irqfd.

ioctl(fd, KVM_IRQ_ACKFD, kvm_irq_ackfd.{flags = 0, fd = eventfd, gsi = $GSI});
ioctl(fd, KVM_IRQ_LINE, kvm_irq_level.{irq = $GSI, level = 1);
/* eventfd notification */
ioctl(fd, KVM_IRQ_LINE, kvm_irq_level.{irq = $GSI, level = 0);

> So why is this an ackfd flag and not an irqfd flag?
> I am guessing because this way you do not need
> an extra ack notifier, but isn't this an internal
> optimization leaking out to userspace?

If irqfd were to setup it's own irq ack notifier for de-assert, we can
guarantee the ordering of de-assert vs eventfd trigger.  So, whoever
does one, needs to do both.  Thanks,

Alex