xl vcpu-set crashes Dom0

xl vcpu-set crashes Dom0

[Andre Przywara]

Hi,

as mentioned earlier today I had some issues with patch 22934, which 
decreases the number of vCPUs for Dom0 in case of cpupool-numa-split.
I now found that this is not directly related to this patch, as it 
triggers also with "xl vcpu-set". So far I tried the following combinations:
c/s 22956 on a machine with 48 physical CPUs:
not restricted dom0:
	xl vcpu-set 0 6  => crashes
         xl vcpu-set 0 47 => crashes
dom0_max_vcpus=6
	xl vcpu-set 0 5  => crashes
         xm vcpu-set 0 5  => crashes

The only message I get is:
root@dosorca:~# xm vcpu-set 0 5
(XEN) Domain 0 crashed: rebooting machine in 5 seconds.

Debugging turned out that for instance the libxl_set_vcpuonline() 
returns without an error, but the machine then crashes afterwards.

Can someone reproduce this? Any ideas?

Regards,
Andre.

-- 
Andre Przywara
AMD-OSRC (Dresden)
Tel: x29712

Re: xl vcpu-set crashes Dom0

[Keir Fraser]

On 28/02/2011 17:23, "Andre Przywara" <andre.przywara@amd.com> wrote:

> Hi,
> 
> as mentioned earlier today I had some issues with patch 22934, which
> decreases the number of vCPUs for Dom0 in case of cpupool-numa-split.
> I now found that this is not directly related to this patch, as it
> triggers also with "xl vcpu-set". So far I tried the following combinations:
> c/s 22956 on a machine with 48 physical CPUs:
> not restricted dom0:
> xl vcpu-set 0 6  => crashes
>          xl vcpu-set 0 47 => crashes
> dom0_max_vcpus=6
> xl vcpu-set 0 5  => crashes
>          xm vcpu-set 0 5  => crashes
> 
> The only message I get is:
> root@dosorca:~# xm vcpu-set 0 5
> (XEN) Domain 0 crashed: rebooting machine in 5 seconds.

Dom0 has oopsed. If you have a synchronous log of dmesg (e.g., via serial)
then you should be able to get an oops message and backtrace from dom0
kernel. My first guess would be this is a bug in the kernel's handling of
cpu hotplug requests via xenstore.

 -- Keir

> Debugging turned out that for instance the libxl_set_vcpuonline()
> returns without an error, but the machine then crashes afterwards.
> 
> Can someone reproduce this? Any ideas?
> 
> Regards,
> Andre.