Re: Windows install problems on Xeon E5-2403 v2

[Ian Campbell <ian.campbell@citrix.com>]

On Thu, 2015-06-11 at 09:01 +0100, Jan Beulich wrote:
> >>> On 11.06.15 at 09:37, <kevin.tian@intel.com> wrote:
> >>  From: Jan Beulich [mailto:JBeulich@suse.com]
> >> Sent: Thursday, June 11, 2015 2:58 PM
> >> 
> >> All,
> >> 
> >> we're seeing recurring but intermittent failures in osstest on just the
> >> two hosts using these particular CPU types. The closest similar ones,
> >> using Xeon E5-2407 v2, do not exhibit this same behavior. We do
> >> note that the latter run at microcode level 0x427, while the problem
> >> ones are only at 0x416, and IanC is in the process of trying to
> >> integrate microcode updates with osstest.

microcode is now being updated automatically in osstest however we are
still seeing incidences of this sort of failure (e.g. [0]). Possibly
less frequent although it is too early to say, either way it still needs
investigation IMHO.

Ian.

[0] http://logs.test-lab.xenproject.org/osstest/logs/58727/test-amd64-amd64-xl-qemuu-win7-amd64/info.html

>  However, since there
> >> doesn't seem to be any matching erratum documented of these CPU
> >> models, we aren't really assuming that a microcode update (0x428
> >> appears to be the newest available) would help here, and hence
> >> we'd like to ask for your assistance in possible further diagnostic
> >> steps, or suggestions towards a resolution.
> >> 
> >> For more context, please see also the mail thread starting at
> >> http://lists.xenproject.org/archives/html/xen-devel/2015-06/msg00522.html.
> > 
> > That's a pretty long thread. Could you summarize what's the
> > error phenomenon?
> 
> Did you at least take a look at the logs of the referenced failure? I
> ask because the failure _is_ the phenomenon; we can't really see
> what may be wrong with the guest. We can only guess that maybe
> it's not receiving (all intended) interrupts anymore.
> 
> Ad-hoc flights run by IanC hint towards "no-apicv" eliminating the
> bad behavior, but didn't give 100% confidence in that.
> 
> > Is the only clue related to microcode so far?
> 
> That's not even a clue, but a wild guess.
> 
> Jan
>