Re: [RFC PATCH] xen/dm: arm: Introudce arm_inject_msi DM op

[Julien Grall <julien@xxxxxxx>]

Hi,

On 20/12/2023 09:13, Mykyta Poturai wrote:
> Hi,
>
> On 20.12.23 02:42, Stefano Stabellini wrote:
> > On Tue, 19 Dec 2023, Julien Grall wrote:
> > > But QEMU should really not need to implement a full ITS. What it needs is a
> > > way to forward the MSI to Xen. That's it.
> > I fully agree with Julien
> >
> >
> > > Stefano, do you have any suggestion how to do this in QEMU?
> > Yes, we just need something like hw/i386/xen/xen_apic.c but for ARM
> It is exactly like xen_apic.c. All this implementation does is getting
> the MSI messages and forwarding them to Xen using the DM op from this patch.
If this implementation is only getting the MSI messages and forwarding 
them, then why is it built on top of the vGICv3 ITS structure?
Shouldn't instead be something that is agnostic to the interrupt 
controller (and even possibly arch agnostic)?
The point here is in Xen context, QEMU doesn't know which interrupt 
controller will be exposed to the guest. QEMU is just used for emulating 
devices and none of this should really be architecture specific. 
Instead, it should be a Xen specific way to inject MSI/IRQ.
Cheers,

--
Julien Grall