Fedora xen - Jan 2007 - HVM guests: Cannot allocate memory

Hi,

my Dom0 is running with Fedora Core 6 and overall I would say it works. But I 
have a problem with HVM guests: I can start them just after system reboot but 
they won''t start after I worked some time. "xm create" just
gives me the error
message "VmError: (12, ''Cannot allocate memory'')".

The error does always appear when my "free" memory (as shown by top)
is lower
than the amount required by my HVM guest. What I do not understand is why Xen is
not able to use memory which is used as cache by the Kernel. It is not uncommon 
for me to have 2-3 GB memory used as cache but only 20-30 MB free. Furthermore I
expected that even SWAP can be used for my Xen guest which are another 2 GB.

Is there a workaround/bugreport for this behavior? Is there any hope it can be 
fixed?

fs

On Fri, Jan 26, 2007 at 11:42:07PM +0100, Felix Schwarz
wrote:
> Hi,
> 
> my Dom0 is running with Fedora Core 6 and overall I would say it works. But
> I have a problem with HVM guests: I can start them just after system reboot
> but they won''t start after I worked some time. "xm
create" just gives me
> the error message "VmError: (12, ''Cannot allocate
memory'')".
> 
> The error does always appear when my "free" memory (as shown by
top) is
> lower than the amount required by my HVM guest. What I do not understand is
> why Xen is not able to use memory which is used as cache by the Kernel. It 
> is not uncommon for me to have 2-3 GB memory used as cache but only 20-30 
> MB free. Furthermore I expected that even SWAP can be used for my Xen guest
> which are another 2 GB.
Unfortunately the ''Cannot allocate memory'' error message is
caused by a bug
in the Xen 3.0.3 hypervisor for HVM. It has nothing todo with the amount of
free memory in either dom0, or the host, and so there''s no easy way to
avoid
it / workaround it.
> Is there a workaround/bugreport for this behavior? Is there any hope it can
> be fixed?
We''ve been trying a number of ways to fix it - the upstream patch from
xen-unstable is too invasive to easily apply to Xen 3.0.3, so we need an
alternate approach. Herbert Xu has just yesterday come up with an alternate
fix which can be done entirely in the userspace tools which we''re
trying
in the RHEL-5 branch of Xen. If it passes testing, then I''ll apply the
patch to FC6 Xen as well & push out an updated RPM at the soonest possible
opportunity (hopefully within next few days).

Regards,
Dan
-- 
|=- Red Hat, Engineering, Emerging Technologies, Boston.  +1 978 392 2496 -=|
|=-           Perl modules: http://search.cpan.org/~danberr/              -=|
|=-               Projects: http://freshmeat.net/~danielpb/               -=|
|=-  GnuPG: 7D3B9505   F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505  -=|

Daniel P. Berrange wrote:
> Unfortunately the ''Cannot allocate memory'' error message
is caused by a bug
> in the Xen 3.0.3 hypervisor for HVM. It has nothing todo with the amount of
> free memory in either dom0, or the host, and so there''s no easy
way to avoid
> it / workaround it.
Thank you very much for your detailled answer :-) At least, the problem is 
already fixed in upstream so I''ll get rid of the bug with FC7 at the
lastest.

fs