On Wed, 2011-08-03 at 10:41 -0400, Nico Kadel-Garcia wrote:
> On Wed, Aug 3, 2011 at 5:12 AM, Matej HALAC <[log in to unmask]> wrote:
> > Hello Gentlemen,
> >
> > we have two HP ProLiant DL380 G6 servers with Intel Xeon E5530
> > processors running libvirt on SL(kernel 2.6.32-131.6.1.el6.x86_64) to
> > host our Linux and Windows servers. (We migrated from Citrix XenServer)
> 
> I like the G6's, they seem to be nice hardware. Do you have those
> ghods-awful Broadcom 10G network cards on them? The ones that try to
> let you split up the 10G into a stack of different slices of
> bandwidth? Those weren't stable enough to use last year, and I doubt
> the drivers or the firmware on those cards has gotten any better.

We have the stock Broadcom Corporation NetXtreme II BCM5709 Gigabit
Ethernet on the machines.
> 
> I'm coming to the harsh conclusion that our favorite upstream vendor's
> "KVM" toolkit is still not ready for production use, especially due to
> the truly awful configuration tool. This was based on horrible
> experience with their 5.5, 5.6, and 6.0 releases. If you have only a
> few hosts to virtualize, why not test Oracle's "VirtualBox" tool
> (which has far better configuration tools and good client
> integration), or VMWare's well supported home editions? Since you seem
> to be using freeware where feasible, such as

I have no doubt that VMWare's vSphere would work like a charm. Like I
said we used XenServer before and everything was fine. So I guess this
must be a specific KVM problem. Maybe with the processor or other server
hardware.

VirtualBox is a fine piece of software but as I understand it its more
of an Desktop solution.

The plan is to have KVM working since we had good experience with it for
testing on our test machines (which were mostly laptops and custom
PC's).
> 
> > The problem is that Windows servers (64bit) running on KVM crash
> > periodically. With the following event log message:
> >        Error code 000000000000003b, parameter1 0000000080000003, parameter2
> > fffff80001039900, parameter3 fffffadfe125fd50, parameter4
> > 0000000000000000.
> 
> Have you done all the updates on both your base system and the
> virtualized systems?

Yeah, everything is updated. We do periodical checks for firmware and
bios updates.
> 
> > The host servers have this in their dmesg that looks suspicious to me:
> >        Performance Events: PEBS fmt1+, Nehalem events, Broken BIOS detected,
> > complain to your hardware vendor.
> >        [Firmware Bug]: the BIOS has corrupted hw-PMU resources (MSR 38d is
> > 330)
> >
> > Also host servers get this message in the logs:
> >        kernel: kvm: 2323: cpu0 unimplemented perfctr wrmsr: 0xc1 data 0xabcd
> >
> > I myself have a ML150 G6 with Intel Xeon E5504 that runs SL6 and libvirt
> > with a Windows server without a hitch.
> >
> > Any advice is appreciated since I tried looking for the solution and
> > nothing helped me.
> 
> You get this on both servers? If you can spare the time, test one of
> the other virtualization technologies.

Yes these messages are present on both machines and also the Windows
image crashes on both as well. The specific machine worked without
problems on my ML150 so I doubt it's a problem with the Guest image.

-- 
HALAC Matej
IT
Canberra - Packard, s.r.o.
Vojtecha Tvrdého 790 / 13
010 01 Zilina
Slovak Republic