Re: [Qemu-devel] QEMU configuration files

[Anthony Liguori]

Fabrice Bellard wrote:
> Hi,
>
> My snapshot of the "object based" QEMU configuration system can be 
> found at http://bellard.org/qemu/patches . I only tried it for x86 
> targets. It is not yet in committable state and comments are welcome !

I've only done a very quick high level review.  I'll try to do a more 
thorough one soon.  My first impressions are very positive.  I think it 
would be better to decentralize class registration but I don't think 
that's a very hard change to make.

> General ideas:
>
> - User preferences and machine definitions are separated. User 
> preferences are in ~/.qemu/config for Unix systems. Machine 
> definitions can override user preferences but I believe it should be 
> the exception.

Ack.  I think this is pretty important.

> - Command line options override the user preferences and machine 
> definitions.
>
> - Machine definitions contain machine parameters and device 
> definitions. Device definitions are used to create new devices not 
> instanciated in the hardcoded machine definition such as PCI and USB 
> devices.
>
> There are many details which need clarification, in particular:
>
> - PCI, IDE, SCSI and buses naming. It is important if we want to be 
> able to dynamically instantiate complicated bus topologies.
> - USB port naming
> - Is it worth specifying board specific network controllers as 
> separate devices (I tried to do that for smc91c111 devices) ? A 
> simpler solution would be to add new machine parameters to do that.
> - It would be logical to define QEMUDevice for every instanciated 
> device and that register_savevm() use QEMUDevice as parameter, but it 
> requires more changes in the code.

Indeed.  QEMUDevice is a good abstraction as it can be also be used to 
introduce per-device locking which will help parallelize things with KVM.

Regards,

Anthony Liguori

> - Is it worth handling class defaults parameters ? I find the current 
> implementation too complicated.
>
> Fabrice.