On 08.07.2013, at 23:06, Anthony Liguori wrote: > Alexander Graf <ag...@suse.de> writes: > >> On 08.07.2013, at 22:08, Anthony Liguori wrote: >> >>> I think we're trying to fit a square peg into a round hole. >>> >>> virtio-mmio is a virtio transport where each device has a dedicated set >>> of system resources. >>> >>> Alex, it sounds like you want virtio-mmio-bus which would be a single >>> set of system resources that implemented a virtio bus on top of it. >> >> Well, what I really want is a sysbus that behaves like PCI from a >> usability point of view ;). > > Which means you need to have (1) a discovery mechanism with a stable > addressing mechanism
That's what dtb usually gives you. > (2) a way to communicate this to the guest from the > host. Not if the host dictates everything. PCI is only complicated because it allows the guest control. If we don't we can have a push-only interface. But I'm not sure we should hold back this patch series based on this. I can try to come up with a bus that can automatically place memory regions and IRQs. Then I can add a virtio-mmio-awesomebus type and show you what I mean ;) For the time being, we can live with a few statically allocated virtio transports I think. As long as we don't promise the user that they're still going to be there in the next version of QEMU. Alex
