Re: [Qemu-devel] Re: [PATCH v2 9/9] Add -kvm option

[Anthony Liguori <anthony@codemonkey.ws>]

>
>>> I still don't like the concept though.  Configuring a second nic would
>>> be different from configuring the first nic, because for the first
>>> you'll modify the default device, the second is added instead.
>>> libvirt folks will hate us for doing this.
>>>       
>
> Having to use -set for configuring the first device of a kind, and
> -device for the second is a bad user interface.  It's made worse by the
> fact that you need -set only for some kinds of devices, namely the kinds
> where QEMU provides a default.
>   

The guest already has a nic, if you want to modify an existing nic, then 
you use -set.  If you want to add a second nic, you use -device.  This 
isn't two difference behaviors.

> I agree with Gerd that we should distinguish between required and
> optional devices.  It's fine to require -set for modifying required
> devices like RTC.  But when I configure my first and second NIC, I don't
> really want to know that I'm actually modifying the first NIC and adding
> the second NIC.
>   

I think what you're arguing for is a truly bare bones machine type where 
there are no assumptions about having a first nic.

But let's look at this from a much higher level.  What is a user 
typically going to want to do?  They're probably going to want to change 
the networking settings globally.  They'll either want to always use tap 
and use the default network device or they're going to want to always 
have two nics because they have to physical devices with separate networks.

A user would either create a new machine type to use for all of their 
machines, or they would modify a global host config to change from slirp 
to tap.  It shouldn't be the common case that they are manually 
specifying -device command line options.  -device is pretty obtuse and 
really is an expert option for things like libvirt and as a placeholder 
for a proper config.

> What about this: give the user a default FOO (for FOO in serial, NIC,
> ...) if he didn't configure one (no matter how, be it -device or some
> legacy option to ask for FOOs).  This way, you ask for your first FOO
> exactly like any other: -device.
>   

You cannot express the concept "give a user FOO if they didn't ask for 
one" in a machine config file.  That forces machine information to be 
baked into qemu which would be unfortunate.  It suggests that you need 
another mechanism to configure what the type of these defaults are.

Regards,

Anthony Liguori