Bug#414040: [Pkg-xen-devel] Bug#414040: kernel: xen_net: Memory squeeze in netback driver.

[Ralph Passgang]

Am Freitag, 9. März 2007 04:07 schrieb Nate Carlson:
> On Thu, 8 Mar 2007, Bob Tanner wrote:
> > Package: xen-hypervisor-3.0.3-1-amd64
> > Version: 3.0.3-0-2
> > Severity: grave
> > Justification: renders package unusable
> >
> > This bug looks to be tracked in upstream:
> >
> > http://bugzilla.xensource.com/bugzilla/show_bug.cgi?id=762
> >
> > Mailing list discussion here:
> >
> > http://lists.xensource.com/archives/html/xen-bugs/2006-11/msg00030.html
>
> IIRC, a workaround is to set a 'dom0_mem' parameter on bootup, so it
> doesn't have to shrink dom0's memory to start the guest domains.

Hi Nate, Bob & all,

dom0_mem often helps, when it comes to problems with memory management, 
especially when pae-kernel & hypervisor are used.

It helps for example when you have buggy hardware, that is not really 100% pae 
(48bit addressing) capabale.

But I haven't watched the phenomena that setting dom0_mem fixes the memory 
squeenze error message that xen brings on some systems. But on all systems I 
saw this error, this never brings the whole system down or crashes something 
or so. It's just ugly.

For me the first questions is, if the reporter (Bob Tanner) really has the 
problem that this makes xen unuseable at all? The bug report itself is rather 
short and only linking to an upstream bug report. Even this upstream bug 
report is not very exact. One is telling about his amd64bit system (2.6.17 
debian kernel with the debian xen-hypervisor which is quite outdated and not 
anymore supported within debian) and someone else responses "I got the same 
with pae" (so obviously not on amd64, but no clue if on debian or even on 
some else distribution). But none of the xen guys reacted on this ticket so 
far.

If this is really a bug that makes xen unusable than I am quite sure this is 
an upstream problem and the xen guys are the right ones to fix this.

Bob, can you give more info on why this makes your system ununsable? Have you 
seen domUs crash or even your whole system? Were you able to write the exact 
kernel oopses down? Have you tried setting "(dom0-min-mem 196)" (or even more 
MB for dom0) and maybe also "(dom0-cpus 1)" in /etc/xen/xend-config.sxp?

regards,
 Ralph