Re: [Qemu-devel] [RFC PATCH v6 0/6] Virtio refactoring.

[Anthony Liguori]

"Michael S. Tsirkin" <address@hidden> writes:

> On Mon, Jan 07, 2013 at 08:02:32PM +0000, Peter Maydell wrote:
>> On 7 January 2013 19:58, Michael S. Tsirkin <address@hidden> wrote:
>> > On Tue, Dec 18, 2012 at 12:30:20PM +0100, KONRAD Frédéric wrote:
>> >> The modifications will be transparent to the user, as we will keep
>> >> virtio-x-pci devices.
>> >
>> > Then what's the point of all this?
>> >
>> > -device virtio-pci,id=transport1 -device virtio-net,bus=transport1
>> >
>> > or
>> >
>> > -device virtio-mmio,id=transport1 -device virtio-net,bus=transport1
>> >
>> > Is simply an insane way to create a network device.
>> 
>> 1. You wouldn't create the virtio-mmio transport on the command line,
>> the machine model does it (it has to because it's a sysbus device
>> and it needs the address/irq lines wiring up properly), so it's just
>>  "-device virtio-net" (and let qemu find the bus automatically)
>
> Bus auto-detection sounds good and would be nice for pci too.
> We had things like model=virtio originally which is pretty close.
> But the issue is, how then do you pass bus specific arguments like pci
> slot? This is what caused us to go the virtio-net-pci route
> to begin with.

PCI is not the same as MMIO here.

virtio-mmio devices are not pluggable.  It makes a lot more sense to
have a virtio-net-pci device.  But it doesn't make much sense to have a
virtio-net-mmio device.

>
>> 2. We shouldn't be making command line simplicity drive how we
>> model devices inside QEMU.
>
> Confused.  I was told that enabling
>  -device virtio-pci,id=transport1 -device virtio-net,bus=transport1
> is the reason we have this patchset.

You were misinformed.

>> If we wanted to do that we should have
>> stuck with the old -net command line arguments which are rather
>> more userfriendly IMHO.
>
> The main thing that confused people with -net was the vlans
> and the need to specify -net twice.
> A good UI would have been e.g. -nic model=virtio,net=user.

That was the original UI.  It was even called -nics.  See
7c9d8e if you're curious.  I was never a fan of the -net syntax.

> But one bad UI does not justify another one.
>
>> If commandline confusion is getting to
>> be a problem with all the -device foo stuff then we should probably
>> fix that at the UI level.
>> 
>> -- PMM
>
> I'd like to see a proposal about how we are going to do this.

(1) Stop conflating internal modeling with UI

(2) Add UI interfaces as appropriate 

It's really that simple.

Regards,

Anthony Liguori

>
> -- 
> MST