Re: [Xen-devel] GRUB Xen PVH chainloading

[Juergen Gross <jgross@xxxxxxxx>]

On 07/01/2019 12:41, Colin Watson wrote:
> Hi,
> 
> I'm working on integrating the recently-merged PVH support for GRUB into
> the Debian packages.  As a result I find myself thinking about how to
> handle the two-stage boot loader scheme that our packages currently
> implement for PV.  I think that it would not be very hard to do this for
> PVH in the manner outlined below, but my x86 asm skills aren't quite up
> to some of the work needed in GRUB.  Assuming that nobody sees any
> obvious holes in this, does anyone fancy giving it a go?

Seems to be a very good idea.

> Background
> ----------
> 
> Around the time PV support was implemented in GRUB 2, we put together a
> scheme to minimise the coupling between the guest configuration file on
> the host and the boot loader configuration in the guest.  The scheme and
> its rationale are described here:
> 
>   
> https://wiki.xen.org/wiki/PvGrub2#Chainloading_guest_pvgrub2_from_domain_0_pvgrub2
> 
> Essentially the same rationale applies to the PVH case: it should be
> possible for the guest to declare its own booting arrangements (though
> of course some hosts may wish to just provide a grub.cfg and handle all
> that on the host side), and this should be decoupled from the GRUB image
> provided by the host as far as possible in order to minimise
> compatibility issues.
> 
> There seems to be no obstacle to this in principle: just as a PV boot
> loader can chainload another one from the guest's disk, so too could a
> PVH boot loader chainload another one from the guest's disk.
> 
> What needs to be done
> ---------------------
> 
> GRUB needs to support chainloading another PVH boot loader.  I think
> this involves observing the existence of the XEN_ELFNOTE_PHYS32_ENTRY
> note and following the machine state rules for the domain builder here:
> 
>   https://xenbits.xen.org/docs/unstable-staging/misc/pvh.html
> 
> (I had a brief go at implementing this, but foundered on my fairly
> minimal understanding of GRUB's relocator/boot code.)

The needed effort should indeed be rather small.

In the moment I have no idea when I'll be able to do it, as I have
plenty of other things to do. In case you want to try it and need some
hints, please feel free to ask (maybe I'm able to give an answer
without having to try to implement it myself ;-) ).

> We need to define a modification to
> https://xenbits.xen.org/docs/unstable-staging/misc/x86-xenpv-bootloader.html
> for PVH boot loaders.  I suggest the obvious: a second-stage PVH boot
> loader should be installed into the guest filesystem as
> /boot/xen/pvhboot-<ARCH>.elf, and otherwise things generally behave the
> same way.  I'd be happy to draft a patch to the protocol specification
> once a proof-of-concept exists.
> 
> The as-yet-unmerged GRUB patch to support the existing PV boot protocol
> (https://salsa.debian.org/grub-team/grub/blob/master/debian/patches/grub-install-pvxen-paths.patch)
> needs to be extended to support the amended protocol.  This is trivial
> given the above.

Would you mind trying to upstream this patch? I have CC-ed Daniel Kiper
one of the grub2 maintainers), as I guess with his Xen skills he will be
the one looking at the patch.


Juergen

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel