Re: [Xen-devel] [PATCH 06/11] xen/mmu: For 64-bit do not call xen_map_identity_early

[Stefano Stabellini <stefano.stabellini@eu.citrix.com>]

On Mon, 20 Aug 2012, Konrad Rzeszutek Wilk wrote:
> On Mon, Aug 20, 2012 at 12:58:37PM +0100, Stefano Stabellini wrote:
> > On Mon, 20 Aug 2012, Ian Campbell wrote:
> > > On Mon, 2012-08-20 at 12:45 +0100, Stefano Stabellini wrote:
> > > > On Fri, 17 Aug 2012, Konrad Rzeszutek Wilk wrote:
> > > > > On Fri, Aug 17, 2012 at 06:41:23PM +0100, Stefano Stabellini wrote:
> > > > > > On Thu, 16 Aug 2012, Konrad Rzeszutek Wilk wrote:
> > > > > > > B/c we do not need it. During the startup the Xen provides
> > > > > > > us with all the memory mapped that we need to function.
> > > > > > 
> > > > > > Shouldn't we check to make sure that is actually true (I am thinking at
> > > > > > nr_pt_frames)?
> > > > > 
> > > > > I was looking at the source code (hypervisor) to figure it out and
> > > > > that is certainly true.
> > > > > 
> > > > > 
> > > > > > Or is it actually stated somewhere in the Xen headers?
> > > > > 
> > > > > Couldn't find it, but after looking so long at the source code
> > > > > I didn't even bother looking for it.
> > > > > 
> > > > > Thought to be honest - I only looked at how the 64-bit pagetables
> > > > > were set up, so I didn't dare to touch the 32-bit. Hence the #ifdef
> > > > 
> > > > I think that we need to involve some Xen maintainers and get this
> > > > written down somewhere in the public headers, otherwise we have no
> > > > guarantees that it is going to stay as it is in the next Xen versions.
> > > > 
> > > > Maybe we just need to add a couple of lines of comment to
> > > > xen/include/public/xen.h.
> > > 
> > > The start of day memory layout for PV guests is written down in the
> > > comment just before struct start_info at
> > > http://xenbits.xen.org/docs/unstable/hypercall/include,public,xen.h.html#Struct_start_info
> > > 
> > > (I haven't read this thread to determine if what is documented matches
> > > what you guys are talking about relying on)
> > 
> > but it is not written down how much physical memory is going to be
> > mapped in the bootstrap page tables.
> 
> Considering that only pvops kernel has this change I think we are ok?

We are OK if we write it down :)
Otherwise it might change in the future and we won't even know what the
correct behavior is supposed to be.