Re: [PATCH 02/12] early PV on HVM

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

On Tue, Jun 08, 2010 at 05:25:52PM +0100, Stefano Stabellini wrote:
> On Tue, 8 Jun 2010, Konrad Rzeszutek Wilk wrote:
> > On Tue, Jun 08, 2010 at 04:55:33PM +0100, Stefano Stabellini wrote:
> > > On Tue, 8 Jun 2010, Konrad Rzeszutek Wilk wrote:
> > > > > > > +	HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
> > > > > > > +
> > > > > > > +	/* Don't do the full vcpu_info placement stuff until we have a
> > > > > > > +	   possible map and a non-dummy shared_info. */
> > > > > > 
> > > > > > Might want to mention where the full vpcu placement is done.
> > > > > 
> > > > > The comment is not accurate, we actually don't do any vcpu_info
> > > > > placement on hvm because it is not very useful there.
> > > > > Better just to remove the comment (I have done so in my tree).
> > > > > 
> > > > > > > +	per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
> > > > > > 
> > > > So.. what is the purpose of the per_cpu(xen_vcpu, 0) then?
> > > > 
> > > 
> > > the vcpu info placement memory area is stored in per_cpu(xen_vcpu_info, cpu);
> > > per_cpu(xen_vcpu, cpu) is just a pointer to that area if it is
> > > available, otherwise it points to the vcpu_info struct in the shared
> > > info page.
> > 
> > I was just wondering why are we doing this when you say:
> > " don't do any vcpu_info placement on hvm because it is not very useful there."
> > 
> > So if it is not useful, why do it?
> > 
> 
> I think Jeremy replied to your question better than me: we still need
> the vcpu_info stuff for the timer and event channels, but we don't need
> it to be at a specific address in kernel memory.

Ok, can you add that comment for the usage of the per_cpu(xen_vcpu,0)
and mention that this is bootstrap code - hence only starting at CPU 0.