Xen vs. GRUB

Xen vs. GRUB

[Thomas Schwinge]

[-- Attachment #1: Type: text/plain, Size: 1542 bytes --]

[Cced to <grub-devel@gnu.org>.]

Hello!

This is about porting GRUB to Xen to allow booting a GNU/Hurd system as a
Xen domU.

GNU/Hurd systems use GNU Mach as a microkernel and have a GRUB
configuration like the following one, which is currently not translatable
to a Xen configuration:

#v+
title Debian GNU/Hurd
kernel (hd0,5)/boot/gnumach.gz \
  root=device:hd0s6
module (hd0,5)/hurd/ext2fs.static \
  --multiboot-command-line=${kernel-command-line} \
  --host-priv-port=${host-port} \
  --device-master-port=${device-port} \
  --exec-server-task=${exec-task} \
  -T typed ${root} $(task-create) $(task-resume)
module (hd0,5)/lib/ld.so.1 \
  /hurd/exec \
  $(exec-task=task-create)
#v-


On Tue, Jan 02, 2007 at 10:21:14PM +0100, Samuel Thibault wrote:
> Thomas Schwinge, le Tue 02 Jan 2007 12:10:49 +0100, a ?crit :
> > On Tue, Jan 02, 2007 at 01:11:05AM +0000, Samuel Thibault wrote:
> > > A crude first ugly patch.
> > 
> > I wondered how you got the thing to boot, wondered if you had also ported
> > GRUB to Xen along the way, but then I saw the patch for
> > `kern/bootstrap.c'.  :-)

[Samuel there hard-coded the above quoted ``module [...]'' lines.]

> ;)
> 
> Yes, the problem is that Xen is a bit Linux-oriented, and as such the
> parameter is called "ramdisk", can ship only one file, and doesn't have
> a command line separate from the kernel's...

Is there a consensus that GRUB (or rather GRUB2) should be ported to
allow them to run in a Xen environment?


Regards,
 Thomas

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 191 bytes --]

Re: Xen vs. GRUB

[Samuel Thibault]

Hi,

Thomas Schwinge, le Wed 03 Jan 2007 13:09:25 +0100, a écrit :
> Is there a consensus that GRUB (or rather GRUB2) should be ported to
> allow them to run in a Xen environment?

I'm not sure grub on Xen would be so useful, since people usually choose
their operating system and parameters directly from Xen configuration
files.  And I actually already contacted Xen people, and they agree that
adding a multiboot-like module abstraction to Xen won't be hard and
would be useful.

Well...  Why not.

Samuel

Re: Xen vs. GRUB

[Ian C. Blenke]

Samuel Thibault wrote:

>Thomas Schwinge, le Wed 03 Jan 2007 13:09:25 +0100, a écrit :
>  
>
>>Is there a consensus that GRUB (or rather GRUB2) should be ported to
>>allow them to run in a Xen environment?
>>    
>>
>
>I'm not sure grub on Xen would be so useful, since people usually choose
>their operating system and parameters directly from Xen configuration
>files.  And I actually already contacted Xen people, and they agree that
>adding a multiboot-like module abstraction to Xen won't be hard and
>would be useful.
>  
>

Xen PV domUs can boot from "pygrub". This is the method used by JailTime 
and rPath to encapsulate kernels and initrd inside the disk images.

So there is already such a "port". Just specify a bootloader in your xen 
config:

    bootloader="/usr/lib/xen/bin/pygrub"

and it's easy enough to override the config file with your own boot entry:

    bootentry="hda1:/boot/vmlinuz-xen,/boot/initrd-xen"

Which does seem a bit silly when you can specify the same thing with 
root=, ramdisk=, and anything else with extra=. Alternatively, there are 
libvirt and xen-tools and other xen config abstrations outside of xm.

Xen HVM domUs, on the other hand, bootstrap whatever MBR is installed on 
the virtual disk. No changes needed here at all.

The next step is HVM with paravirt_ops/VMI, which might actually show up 
some day for Xen (not there yet). Guests would still run in an HVM jail 
but are aware of the hypervisor and don't need to interact with the QEMU 
provided virtual hardware (think PV drivers).

Also, kvm has paravirt_ops now, as of a couple of days ago (thanks to 
Ingo). With l-hype and kvm, xen is quickly becoming moot anyway.

How does this affect grub and other bootstrapping? We very well might be 
talking about paravirt_ops/VMI and/or PV drivers at some point in 
addition to vanilla int13 real mode devices (though the virtual BIOSes 
will likely take care of this anyway).

- Ian C. Blenke <ian@blenke.com>

Re: Xen vs. GRUB

[Samuel Thibault]

Hi,

Ian C. Blenke, le Mon 08 Jan 2007 13:56:32 -0500, a écrit :
> Xen PV domUs can boot from "pygrub". This is the method used by JailTime 
> and rPath to encapsulate kernels and initrd inside the disk images.
> 
> So there is already such a "port". Just specify a bootloader in your xen 
> config:
> 
>    bootloader="/usr/lib/xen/bin/pygrub"
> 
> and it's easy enough to override the config file with your own boot entry:
> 
>    bootentry="hda1:/boot/vmlinuz-xen,/boot/initrd-xen"

Mmm, how does the domU kernel get the store/console MFN/port and such
info that is usually in the xen boot_info structure?

Samuel

Re: Xen vs. GRUB

[Yoshinori K. Okuji]

On Wednesday 03 January 2007 13:09, Thomas Schwinge wrote:
> Is there a consensus that GRUB (or rather GRUB2) should be ported to
> allow them to run in a Xen environment?

Sorry for my late response. Personally, I think it is a good idea to port GRUB 
2 to Xen. Xen is just a firmware from GRUB's point of view, so it should not 
be very difficult to do.

In addition, I would like to see GRUB 2 ported to more platforms, such as 
LinuxBIOS as well as Xen.

Okuji

Re: Xen vs. GRUB

[Tristan Gingold]

On Fri, Jan 26, 2007 at 07:46:59PM +0100, Yoshinori K. Okuji wrote:
> On Wednesday 03 January 2007 13:09, Thomas Schwinge wrote:
> > Is there a consensus that GRUB (or rather GRUB2) should be ported to
> > allow them to run in a Xen environment?
> 
> Sorry for my late response. Personally, I think it is a good idea to port GRUB 
> 2 to Xen. Xen is just a firmware from GRUB's point of view, so it should not 
> be very difficult to do.
Humm, you have to write a few device drivers...

What is the point in porting grub to xen ?  Xen directly loads the kernel and
Xen comes with pygrub (a grub clone in python) which allows the user to
choose kernel and set parameters.  (Although I have never tried it).

Tristan.