On Mon, May 23, 2011 at 11:44:38AM +0100, Ian Campbell wrote:
On Tue, 2011-05-17 at 16:59 +0100, Ian Campbell wrote:
The following implements all your feedback (I hope). I have squashed it down into a single commit which supports direct boot.
BTW, in general, the above patch looks okay to me.
I've gotten all the PCI setup and ACPI stuff etc etc working but, frankly, the patch to SeaBIOS is getting to be pretty enormous and intrusive.
Is that due to incompleteness / innacuracies in the current SeaBIOS code, or due to requirements specific to Xen?
If you have test code, I'd be curious to see a patch on the mailing list - it may help later to understand the use cases for SeaBIOS.
The more I look at it the more I am coming to the conclusion that it would be better to have hvmloader setup all this platform level stuff and pass details onto SeaBIOS as necessary, following the model used with coreboot->SeaBIOS rather than the emulator's way of doing things. hvmloader basically already fulfils the same role for Xen HVM guests as coreboot does for physical hardware so I think that makes a certain amount of sense.
I'm okay with that approach as well - there are pros and cons to each method. (At various points, it's been discussed whether SeaBIOS should generate ACPI tables for coreboot, and it has also been discussed if QEmu should just pass in ACPI tables to SeaBIOS..)
-Kevin