Re: [PATCH 4 of 4] xen/pagetables: Document pt_base inconsistency when running in COMPAT mode

[Konrad Rzeszutek Wilk <konrad@kernel.org>]

On Wed, Aug 22, 2012 at 10:21:10AM +0100, Ian Campbell wrote:
> On Tue, 2012-08-21 at 21:08 +0100, Konrad Rzeszutek Wilk wrote:
> > # HG changeset patch
> > # User Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
> > # Date 1345579709 14400
> > # Node ID b992f8e613a3401b9ddd140ce436c840d412beb7
> > # Parent  74bedb086c5b72447262e087c0218b89f8bc9140
> > xen/pagetables: Document pt_base inconsistency when running in COMPAT mode.
> > 
> > c/s 13257 added the COMPAT mode wherein a 64-bit hypervisor can
> > run with a 32-bit initial domain. One of the things it added was
> > that the pt_base is offset by two pages. This inconsistency
> > is only present in the COMPAT mode so lets document it.
> > 
> > Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
> > 
> > diff -r 74bedb086c5b -r b992f8e613a3 xen/include/public/xen.h
> > --- a/xen/include/public/xen.h	Tue Aug 21 16:08:29 2012 -0400
> > +++ b/xen/include/public/xen.h	Tue Aug 21 16:08:29 2012 -0400
> > @@ -663,7 +663,7 @@ typedef struct shared_info shared_info_t
> >   *      c. list of allocated page frames [mfn_list, nr_pages]
> >   *         (unless relocated due to XEN_ELFNOTE_INIT_P2M)
> >   *      d. start_info_t structure        [register ESI (x86)]
> > - *      e. bootstrap page tables         [pt_base, CR3 (x86)]
> > + *      e. bootstrap page tables         [pt_base, CR3 (x86)] *1
> >   *      f. bootstrap stack               [register ESP (x86)]
> >   *  4. Bootstrap elements are packed together, but each is 4kB-aligned.
> >   *  5. The initial ram disk may be omitted.
> > @@ -678,6 +678,9 @@ typedef struct shared_info shared_info_t
> >   *
> >   *  NOTE: The initial virtual region (3a -> 3f) are all mapped by the initial
> >   *  pagetables [pt_base, CR3 (x86)].
> > + *
> > + *  *1: When booting under a 64-bit hypervisor with a 32-bit initial domain
> > + *  it is offset by two pages (pt_base += PAGE_SIZE*2).
> 
> "it" here being the page which you would have to load into cr3?

No. The pt_base value.
> 
> What, if anything, ends up in the other two pages?

The L3 and L1. The pt_base has incorrect data and points to second L1 instead
of the L3.

Its laid out like this: [L3] [L1] [L1] ... [L2]
                         a    b    c

and pt_base has the phys_addr of 'c' instead of 'a'.