Re: [Qemu-devel] [PATCH 3/6] virtio-console: Add a virtio-console bus, support for multiple ports

[Amit Shah <amit.shah@redhat.com>]

On (Wed) Oct 07 2009 [11:51:14], Gerd Hoffmann wrote:
> On 10/07/09 11:25, Amit Shah wrote:
>>>    -device virtio-serial-bus,chardev=<name>
>>>
>>> automatically creates a virtioconsole with port=0 and chardev=<name>  on
>>> the newly created bus.
>>
>> Hm, this looks weird. Because on one hand we're talking about decoupling
>> the char driver from the core (virtio-serial-bus) and here we're
>> actually attaching a char driver to the bus.
>
> We don't actually attach the chardev to the bus though.  It is just  
> passed through to the auto-created console port #0.

Yes, but it's unituitive... and weird.

> But it looks a bit weird indeed.  I'm open to better suggestions to  
> address the "port #0 must be console for backward compatibility reasons"  
> issue.

Hm, me too. I'll think about this more.

>>> Keeping -virtioconsole for backward compatibility is easy, it would
>>> basically create a chardev with a virtio<nr>  label as it does today,
>>> then create virtio-serial-bus with chardev=virtio<nr>.
>>
>> I prefer to remove the -virtioconsole argument because we won't be able
>> to specify the bus that's to be attached to.
>
> Oh, I thought it would create a new virtio-serial-bus (plus auto-created  
> port0 console) unconditionally.  Just do enougth to keep existing users  
> of the switch working.
>
> If you want new features (i.e. two consoles ports attached to one  
> virtio-serial-bus device) you must use the new syntax.

So it's better overall to drop the old syntax altogether, right? It
could get easily confusing otherwise.

We can easily end up having:

-virtioconsole <chardev>
<auto-creates a bus and attaches a console port to it>

-device virtio-serial-pci,id=blah
<a second bus>

-device virtport,bus=blah.0

<and no way to connect a device to the bus that got created by
virtioconsole>

		Amit