Re: [PATCH] x86: use pgd accessors when cloning a pgd range.

[Ian Campbell <Ian.Campbell@eu.citrix.com>]

On Wed, 2010-10-27 at 18:51 +0100, Jeremy Fitzhardinge wrote:
> On 10/27/2010 10:42 AM, H. Peter Anvin wrote:

> > Ian's message just mentioned "a failure" and never described in any way what
> > kind of "failure" it was.
> 
> It would be a pagefault from Xen preventing a direct write to the pgd
> level of an active pagetable.  At the point in setup_arch() where it
> does the first clone_pgd_range() we're already running on swapper_pg_dir
> and the copy from initial_page_table is outright wrong.

I'd missed that aspect, yes the contents of initial_page_table are
wrong. I'm not sure now how my patch to clone_pgd_range even made a
difference...

> As Ian suggests, we could switch Xen to use initial_page_table at boot
> then move to swapper_pg_dir in the same way native does.

Accidentally did that in private mail, for everyone else the gory
details are:

> xen_setup_kernel_pagetable operates on initial_page_table instead of
> swapper_pg_dir. We do not pin initial_page_table apart from the
> implicit
> one from writing it to cr3 (it is necessarily r/o though).
> 
> So we enter setup_arch running on initial_page_table and with
> swapper_pg_dir mapped r/w, which is how it looks on native too. So at
> this point the:
>         clone_pgd_range(swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
>                         initial_page_table + KERNEL_PGD_BOUNDARY,
>                         KERNEL_PGD_PTRS);
> works just fine and we reach:
>       load_cr3(swapper_pg_dir);
> 
> Then in __xen_write_cr3 we notice the first attempt to switch to
> swapper_pg_dir and only at that point do we make it r/o and pin it.
> When
> we then actually do the switch to swapper_pg_dir that releases the
> implicit pin on initial_page_table so we can make it r/o again.
> 
> The later on we reach the:
>        clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
>                         swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
>                         KERNEL_PGD_PTRS);
> which will succeed because initial_page_table is r/w again.
> 
> We don't care about the pin on initial_page_table because we should
> never need it again.

Ian.