On Mon, 2005-03-21 at 23:01 +0000, Tim Deegan wrote:
> On Mon, Mar 21, 2005 at 09:57:39PM -0000, Ian Pratt wrote:
> > As I recall, mbootpack uses Linux's x86 BIOS code too, so this shouldn't
> > be an issue.
> mbootpack does have a number of changes to the linux boot code, in order
> to prepare a Multiboot environment; this might be a bit heavyweight as I
> don't imagine is has much use apart from for Xen.
> > Is there a downside I haven't spotted?
I think it boils down to an argument of whether or not the boot_params
approach used between setup.S and vmlinux can be supported with minimal
code. As I said before, and as you can see from the code I posted, I
don't think it is very difficult to do. I would argue that the
complexity of Xen handling boot_params is lower than that of assembly
code in mbootpack's setup.S to setup a multiboot environment.
Additionally, if you are willing to take the boot_param's as provided by
Linux's setup.S, then you also inherit all of the BIOS probe data that
setup.S performed. Currently, only the E820 data is used, but
conceivably one could export some of this data to dom0, as it doesn't
get to run the BIOS probes as a bare-metal Linux would. (I don't how
important such data is though.)
> The xenlinux image, command-lines and initrd can't be altered at boot
> time; it looks like Michal's approach has this problem too.
Michal Ostrowski <mostrows@xxxxxxxxxxxxxx>
Description: This is a digitally signed message part