On Wed, May 01, 2013 at 07:53:49AM +0100, Anup Patel wrote:
On Wed, May 1, 2013 at 7:37 AM, Rusty Russell rusty@rustcorp.com.au wrote:
Alexander Graf agraf@suse.de writes:
There are not device specific registers in virtio-console. Virtio-console lives behind a virtio bus which doesn't know what these registers are.
You're not going to make coherent arguments without reading that actual patch we're discussing. And you're going to just waste everyone else's time.
Even if you shove it into config space,
Which is what the patch does...
it'd be broken because config access happens without intercepts on some platforms.
But people keep assuming it, which is one reason virtio_ccw switched from passive lguest-style config to active. Only lguest and the old kvm virtio use it, and lguest can change.
An emergency output is a reasonable idea, and this is a reasonable implementation. The question is practical: will it be used? Because we don't implement reasonable ideas which aren't going to be used.
We are already using it for Guest bring-up on APM ARM64 board using KVMTOOL and so far it has been really helpful in debugging KVM ARM64 guest issues.
Also, this is a very useful debugging method for Guest bring-up on other architectures supported by KVM and other Hypervisors that support VirtIO.
FWIW: I've hacked together a small set of patches allowing us to use the x86 ioport peripherals on arm/arm64 in kvmtool. This means you can use the 8250 for earlyprintk on the latter, and even as the main console if you like since it has a device-tree binding.
I'll post the patches as an RFC to the kvm list.
Will