Comment 26 for bug 1304001

Hi Andrew,

sorry about the dyndbg. It all lokked like it would be tied to that but right now I am not sure it is. I need to check the code again. If anything else fails we can do a special kernel. I have the same issue as Boris in that I am not able to cause the same problem on HVM guests locally. I am using an Ubuntu base install (Xen-4.2.2 + dom kernel 3.8). xend-config is set to allow ballooning.
Now the question is how memory for a guest is configured on the host. I can see the issue on a c3.large instance but we have no way of knowing anything outside the guest.

One note, too. There is one other memory related message in the guests dmesg:
[ 0.000000] NUMA: Warning: invalid memblk node 0 [mem 0x100000000-0x0fffffff]

I vaguely remember NUMA related problems at some point. But I cannot remember whether this was something that needed fixes in the hypervisor, the dom0 kernel or the guest kernel. Or combinations of that.