Re: [PATCH RFC] kvm: enable irq injection from interrupt context

[Gleb Natapov <gleb@redhat.com>]

On Thu, Sep 16, 2010 at 03:38:28PM +0200, Michael S. Tsirkin wrote:
> On Thu, Sep 16, 2010 at 03:14:11PM +0200, Avi Kivity wrote:
> >  On 09/16/2010 02:57 PM, Michael S. Tsirkin wrote:
> > >>
> > >>  >  >  If you want to split parts that asserts irq and de-asserts it then we
> > >>  >  >  should have irqfd that tracks line status and knows interrupt line
> > >>  >  >  polarity.
> > >>  >
> > >>  >  Yes, it can know about polarity even though I think it's cleaner to do this
> > >>  >  per gsi. But it can not track line status as line is shared with
> > >>  >  other devices.
> > >>  It should track only device's line status.
> > >
> > >There is no such thing as device's line status on real hardware, either.
> > >Devices do not drive INT# high: they drive it low (all the time)
> > >or do not drive it at all.
> > >
> > 
> > That's just an implementation detail.  Devices either assert INT# or
> > they do not.  Tying the wires together constitutes an AND gate.
> > This gate has to be modelled somewhere, currently it's in qemu's pci
> > emulation.
> 
> Right. kvm in kernel has this as well, we need to keep this in
> kvm kernel if we want to support level with irqfd.
> Where it does not belong is individual devices: these
> should be able to assert INTx multiple times
> and it should have no effect, as per spec.
Assert_INTx/Deassert_INTx you mentioned are internal PCI thing. What KVM
sees logically is status of the line between pci controller and irq
chip. We do not emulate PCI inside kernel, but I agree that kernel should
handle multiple asserts without de-assert in the middle and, in fact, it does.
But the thread started with you trying to optimize this non-optimal
device behaviour and I am saying that the fix should be elsewhere. Namely in
irqfd.

--
			Gleb.