Re: [Qemu-devel] [Xen-devel] [XEN][RFC PATCH V2 12/17] xl: Add interface to handle qemu disaggregation

[Ian Campbell]

On Fri, 2012-08-24 at 13:56 +0100, Julien Grall wrote:
> On 08/23/2012 02:30 PM, Ian Campbell wrote:
> > On Wed, 2012-08-22 at 13:31 +0100, Julien Grall wrote:
> >    
> >> This patch modifies libxl interface for qemu disaggregation.
> >>      
> > I'd rather see the interfaces changes in the same patch as the
> > implementation of the new interfaces.
> >
> >    
> >> For the moment, due to some dependencies between devices, we
> >> can't let the user choose which QEMU emulate a device.
> >>
> >> Moreoever this patch adds an "id" field to nic interface.
> >> It will be used in config file to specify which QEMU handle
> >> the network card.
> >>      
> > Is domid+devid not sufficient to identify which nic?
> >    
> Is the user can specify or find devid easily ?
> I added "id" because, I would like that the user
> can identify without any problem a network
> interface.

At the libxl level the libxl_device_nic struct has a devid in it.

That's not to say that xl can't add a layer of naming and indirection on
top.

> >> @@ -246,6 +246,20 @@ libxl_domain_sched_params = 
> >> Struct("domain_sched_params",[
> >>       ("extratime",    integer, {'init_val': 
> >> 'LIBXL_DOMAIN_SCHED_PARAM_EXTRATIME_DEFAULT'}),
> >>       ])
> >>
> >> +libxl_dm_cap = Enumeration("dm_cap", [
> >> +    (1, "UI"), # Emulate all UI + default device
> >>      
> > What does "default device" equate too?
> >    
> The following devices:
>     - i440fx
>     - piix3
>     - piix4
>     - dma
>     - xen apic
>     - xen platform

So this is more like "CORE" than "UI"?

Is there a reason why UI (which I guess means the VGA, spice and VFB
devices?) are required to be in the same emulator as these?

> >> +    (2, "IDE"), # Emulate IDE
> >> +    (4, "SERIAL"), # Emulate Serial
> >> +    (8, "AUDIO"), # Emulate audio
> >> +    ])
> >> +