Re: Start xen on UEFI system with grub

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

On Fri, Nov 02, 2012 at 02:11:22PM +0000, Jan Beulich wrote:
> >>> On 02.11.12 at 14:52, Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> wrote:
> > On Thu, Nov 01, 2012 at 07:57:30PM +0000, Jan Beulich wrote:
> >> >>> Stefano Stabellini <stefano.stabellini@eu.citrix.com> 11/01/12 4:50 PM >>>
> >> >On Tue, 11 Sep 2012, Jan Beulich wrote:
> >> >> >>> On 11.09.12 at 15:17, Fabio Fantoni <fantonifabio@tiscali.it> wrote:
> >> >> > xen.efi not compile when we build xen on Wheezy and probably is not 
> >> >> > possible boot with lvm volume, fallback options ecc...
> >> >> 
> >> >> Just get a suitable tool chain installed then.
> >> >> 
> >> >> > UEFI with grub2 seem the best option but with xen hypervisor seem not 
> >> >> > load efi variable.
> >> >> 
> >> >> Sure, because only xen.efi has the code to deal with such.
> >> >> 
> >> >> (Btw., I assume you aren't aware that any boot manager
> >> >> whatsoever, other than the one coming with EFI, is sort of
> >> >> bogus under EFI?)
> >> >
> >> >Unfortunately it looks like more and more people are of the opinion that
> >> >grub2 is useful even on UEFI firmware.
> >> >For example, it is pretty clear that Ubuntu is still going to boot Xen
> >> >via grub2 no matter how the firmware looks like.
> >> 
> >> Then they need to be tought how to do it properly, the more that Linux 
> > itself
> >> now is capable of booting without a boot loader (other than the EFI one).
> >> 
> >> >What would it take to make Xen work properly with grub2 on a UEFI firmware?
> >> 
> >> I can't be that difficult.
> > 
> > I believe the only issue was the check to do the  Multiboot-E820 parsing
> > before the E801 parsing? Perhaps that has already been done?
> 
> That's in all 4.x trees already. But no - the other important aspect
> is that without retrieving/passing on the proper EFI table pointers,
> on a legacy free system Xen and Dom0 won't be able to locate the
> ACPI tables.

Ah yes (<smacks his head - this is what this thread is about after all>).

I think the work-around could be to use the kexec introduced arguments:

"acpi_rsdp=<physical address of rsdp>"

> 
> Jan
>