PAE requirement in config file for HVM guests

PAE requirement in config file for HVM guests

[Jan Beulich]

Could anyone explain why it was necessary in 3.0.3 to require config files
for 32-bit/PAE or 64-bit x86 hvm guests on respective hypervisors to have
an explicit 'pae = 1', whereas in 3.0.2 this wasn't needed? It would seem
that the emulated CPUID feature flag should, as in 3.0.2, just be
suppressed if the hypervisor really can't support PAE (i.e. 32-bit non-PAE
build).

Thanks, Jan

Re: PAE requirement in config file for HVM guests

[Keir Fraser]

On 22/11/06 08:36, "Jan Beulich" <jbeulich@novell.com> wrote:

> Could anyone explain why it was necessary in 3.0.3 to require config files
> for 32-bit/PAE or 64-bit x86 hvm guests on respective hypervisors to have
> an explicit 'pae = 1', whereas in 3.0.2 this wasn't needed? It would seem
> that the emulated CPUID feature flag should, as in 3.0.2, just be
> suppressed if the hypervisor really can't support PAE (i.e. 32-bit non-PAE
> build).

It's only to allow pae to be masked off. I think that you can specify pae=1
always and it won't be advertised to the guest if the host doesn't support
it. If this isn't the case it should be, and I'd take a patch to make it so.
:-)

Oh, and xend should default pae=1 if it doesn't do so already.

 -- Keir