Re: Debian linux-image-2.6.32-4-xen-amd64 2.6.32-11 doesn't boot with > 4 GiB; resets immediatelly, no log messages

[Jeremy Fitzhardinge <jeremy@goop.org>]

On 05/08/2010 08:46 AM, Thomas Schwinge wrote:
>> From: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
>> Subject: [PATCH] xen/core: don't bother trying to free pages beyond the ones Xen gave us
>>
>> Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
>>     
> I put this on top of Debian linux-2.6 2.6.32-11, which is based on
> f64df18aae5ab07b44bdcc2334cf0044ef46320c of
> git://git.kernel.org/pub/scm/linux/kernel/git/jeremy/xen.git.  Please
> tell me if I shall put it on top of another version.
>
> The patch does mitigate / hide the problem, as the 100000000 to 120000000
> pages are now chopped off (which is not what you intended to do, as I
> understood it):
>   

No, it looks like its working as expected, though its perhaps not ideal
if you want to give dom0 all host memory.  Xen is giving dom0 ~4G of
pages, from 0-4G in pseudo-physical space.  Unfortunately that overlaps
with the PCI window which is somewhere in the 3-4G range.  While in
theory they're in distinct address spaces (pseudo-phys vs machine phys),
it is too hard to deal with that (it is not a concept that Linux really
understands).  The hardware avoids the PCI hole by remapping the memory
to above 4G, but that's not visible to dom0.

The end result is that some memory Xen gave dom0 in the PCI hole can't
be used, but rather than wasting it we just free it back to Xen.  It
will get used as soon as you start up a domain (or, in principle, you
could balloon it back into dom0, but we don't currently support that).

    J