[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-users] domU kernel crash after live migration on Debian 7.5

On Tue, 2015-01-27 at 16:47 +0100, John Naggets wrote:

> I don't know what cpuid levelling is, so I would just answer here:
> whatever is default in Xen.

The default is no.

> My guess here would be the older processor which actually never popped
> up into my mind. I thought it would be possible to live migrate live
> between any type of processors as long as the architecture is the
> same.

Once a feature has been exposed to an OS (likely at boot time) then if
you move another processor it also needs to be able to provide the same
feature, otherwise the instruction will fault (since the CPU knows
nothing about it).

>  Do you have any pointers or documentation about how "much" processors
> are allowed to differ for live migration still being able to do its
> job correctly?

Ideally they would be identical, or at least differ only in the
stepping. If this isn't the case, e.g. the model or family differs, then
any features which the source CPU has and the target CPU lacks will need
to be hidden from the guests somehow.

http://xenbits.xen.org/docs/unstable/man/xl.cfg.5.html describes how you
can use the cpuid directive in your guest cfg to hide features in the
cpuid registers from the OS on a guest by guest basis (YMMV may vary
with Xen 4.1 and Ganetti, not sure to what extent they supply/expose
this respectively).

http://xenbits.xen.org/docs/unstable/misc/xen-command-line.html also
shows options for masking cpuid at the host level, again I'm not sure to
what extent this worked with Xen 4.1 though.

For some features it is also possible to configure the guest kernel not
to use them (e.g. no-xsave on the Linux command line).

It's XenServer specific but http://support.citrix.com/article/CTX127059
has some useful background. The pool.other-config:cpuid_feature_mask
setting referenced there corresponds to the cpu_mask_* in the command
line ref (I think, not 100% sure though)

>  For reference here is Intel's comparison of my two CPUs:
> http://ark.intel.com/compare/75269,48768

http://en.wikipedia.org/wiki/RdRand says RDRAND is also known as "Secure
Key" and that ark.intel.com page shows that Ivy Bridge has it and
Westmere doesn't.


Xen-users mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.