[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] Re: [Xen-users] cpuid masking on Xen 4.1
On 24/03/16 08:48, Charles Koprowski wrote: > >>> I'm not really familiar with with cpuid leveling, but I think there's >>> indeed something wrong. I would expect that having cpu_has_cpuid_faulting >>> means we would do the masking somewhere else (in the fault handler), but I >>> cannot find any usages of opt_cpuid_mask_* anywhere outside of the intel >>> and amd cpu initialization files. >> If you look in the XenServer patch queue >> (https://github.com/xenserver/xen-4.6.pg/) at a changeset older than >> b39aebaf663395714ffcff19e9707500a1580d3f, you will find >> xen-apply-cpuid-mask-to-cpuid-faulting.patch which causes the cpuid >> masks to apply to faulting as well. >> >> All of this however has been superseded by my cpuid improvement series >> which is currently at v3 and expected to land in Xen 4.7. >> >> What are you actually trying to do? > I'm running a cluster of 4 hosts that run Xen 4.1 and share a > clustered LVM group. They all have Intel Xeon E5-4620 CPU (Dell PE > R820) and run Ubuntu 12.04 > > I'm currently extending the cluster with a brand new Dell PE R930 > which have Intel Xeon E7-4850 v3 (unfortunately R820 serie has been > discontinued). > > So I'm trying to "downgrade" the R930 CPU in order to be able to use > live migration. Currently when I initiate a live migration from the > R930 to one of the other hosts, the whole machine freeze without > raising any message. I thought that only the guest OS would crash > instead... Please can you be more specific. What is "the whole machine"? Do you mean the R930 host, an R820 host, or the virtual machine under migrate? > Do you think I can run a much recent version of Xen on the R930 and be > able to use live migration with the rest of the cluster ? Live migration will not function trying to migrate from a newer version of Xen to an older version. You must match your versions of Xen across the pool. > I also tried setting the cpuid option at guest level configuration but > didn't manage to mask all unwanted CPU features. Feature levelling is tricky. I have spent the better part of a year developing a robust solution to the problem. If you are interested in some of the details, http://xenserver.org/discuss-virtualization/virtualization-blog/entry/cpu-feature-levelling-in-dundee.html Your chances of getting this to work with Xen 4.1 is zero. If my work gets into Xen 4.7, your chances of making that work are ok, but by no means certain. xl/libxl is not a multi-host toolstack, and lacks a lot of the considerations required. ~Andrew _______________________________________________ Xen-users mailing list Xen-users@xxxxxxxxxxxxx http://lists.xen.org/xen-users
|
Lists.xenproject.org is hosted with RackSpace, monitoring our |