shadow_translated pv guests?

shadow_translated pv guests?

[Gerd Hoffmann]

  Hi,

Are there any?  I've seen shadow mode fixes in unstable recently, but
trying to boot the vmlinux kernel built from the sparse tree with
features=auto_translated_physmap fails.  Is this supposed to work?  Or
you are using other auto-translated guest kernels?

cheers,
  Gerd

-- 
Gerd Hoffmann <kraxel@suse.de>

Re: shadow_translated pv guests?

[Tim Deegan]

At 14:36 +0100 on 13 Dec (1166020610), Gerd Hoffmann wrote:
> Are there any?  I've seen shadow mode fixes in unstable recently, but
> trying to boot the vmlinux kernel built from the sparse tree with
> features=auto_translated_physmap fails.  Is this supposed to work?  Or
> you are using other auto-translated guest kernels?

At the moment, only the paravirt-ops kernels use translated mode, and
they don't do explicit MMU operations to update their pagetables. 

Cheers,

Tim.

Re: shadow_translated pv guests?

[Gerd Hoffmann]

Tim Deegan wrote:

> At the moment, only the paravirt-ops kernels use translated mode, and
> they don't do explicit MMU operations to update their pagetables. 

The paravirt-ops kernels (built from rusty's patch queue) fail to boot
for me (using a ttylinux ramdisk).  No matter whenever
CONFIG_XEN_SHADOW_MODE is enabled or not.  The shadow kernel dies quite
early at boot.  The non-shadow seems to fail execute init.

Hmm ...

  Gerd

-- 
Gerd Hoffmann <kraxel@suse.de>

Re: shadow_translated pv guests?

[Jeremy Fitzhardinge]

Tim Deegan wrote:
> At the moment, only the paravirt-ops kernels use translated mode, and
> they don't do explicit MMU operations to update their pagetables. 
>   

Actually the paravirt_ops kernel does support full pv-mmu operation; its
use of shadow was a crutch from before I'd got all that going.  The main
reason pv_ops shadow mode is interesting at the moment is that 2.6.20
has the core of paravirt ops, but none of the mmu stuff, so we'll need
Xen to operate in shadow mode if we're to get it into 2.6.20.

That said, xen-unstable does still seem to have problems with shadow pv
guests; I've got !PAE to work, but PAE crashed Xen last time I tried.

    J

Re: shadow_translated pv guests?

[Jeremy Fitzhardinge]

Gerd Hoffmann wrote:
> The paravirt-ops kernels (built from rusty's patch queue) fail to boot
> for me (using a ttylinux ramdisk).  No matter whenever
> CONFIG_XEN_SHADOW_MODE is enabled or not.  The shadow kernel dies quite
> early at boot.  The non-shadow seems to fail execute init.
>
> Hmm ...
>   

When did you last try? !SHADOW_MODE works for me, but there's been quite
a lot of flux in there over the last couple of days.  Chris Wright can't
make it work either, so I think there's something .config dependent in
there.

    J

Re: shadow_translated pv guests?

[Gerd Hoffmann]

Jeremy Fitzhardinge wrote:
> When did you last try? !SHADOW_MODE works for me, but there's been quite
> a lot of flux in there over the last couple of days.  Chris Wright can't
> make it work either, so I think there's something .config dependent in
> there.

yesterday.  "make defconfig" and then the minimum set of changes needed
(pc not genericarch, smp off, paravirt everything on).

cheers,
  Gerd

Re: shadow_translated pv guests?

[Gerd Hoffmann]

  Hi,

> When did you last try? !SHADOW_MODE works for me, but there's been quite
> a lot of flux in there over the last couple of days.  Chris Wright can't
> make it work either, so I think there's something .config dependent in
> there.

I'm trying to verify the rewritten domain builder works ok for shadow
translated guests, but for that I need a working environment in the
first place, right now I didn't manage to boot a single shadow
translated pv guest even with the old builder (3.0.4-rc1) ...

Related thing:  I've noticed even the shadow_translated kernels get a
p2m map by the domain builder, is it needed/used for something?

cheers,
  Gerd

-- 
Gerd Hoffmann <kraxel@suse.de>

RE: shadow_translated pv guests?

[Ian Pratt]

> > When did you last try? !SHADOW_MODE works for me, but there's been
quite
> > a lot of flux in there over the last couple of days.  Chris Wright
can't
> > make it work either, so I think there's something .config dependent
in
> > there.
> 
> I'm trying to verify the rewritten domain builder works ok for shadow
> translated guests, but for that I need a working environment in the
> first place, right now I didn't manage to boot a single shadow
> translated pv guest even with the old builder (3.0.4-rc1) ...
> 
> Related thing:  I've noticed even the shadow_translated kernels get a
> p2m map by the domain builder, is it needed/used for something?

It could in principle be used by driver domains, but could be optional.

Ian


> --
> Gerd Hoffmann <kraxel@suse.de>
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel