kvm guest hangs [was Re: Xen Dom0 [was Re: Announcing Fedora 11 Alpha (blink)]]
debian at herakles.homelinux.org
Fri Feb 6 10:10:49 UTC 2009
Mark McLoughlin wrote:
> On Fri, 2009-02-06 at 18:55 +0900, John Summerfield wrote:
>> Mark McLoughlin wrote:
>>> Try booting the guest with clocksource=acpi_pm on the kernel command
>>> line. If that helps, it's this bug:
>>> It's an issue with KVM's paravirt clock. We have a workaround in the
>>> rawhide host kernel which disables kvmclock support on affected hosts. A
>>> real fix is in progress upstream.
>>> The F-10 branch has the workaround, but only for the 2.6.29 builds. I
>>> don't think we have it in the 2.6.27 updates (i.e. the
>>> private-fedora-10-2_6_27 branch).
>>> Chuck, could we get linux-2.6-kvmclock-unsync-tsc-workaround.patch
>>> pulled into F-10?
>> I'm fully virtualised. I think. What's this about a paravirt clock in KVM?
> KVM uses hardware virtualization support to run guests, yes, but it also
> allows guests to use paravirtualization for some performance critical
> operations - so e.g. pv block/network drivers, pv clock and pv MMU.
My debian guest uses tsc for its clock source (I just looked under sys,
having read that bug report and got a clue). It has none of the pv stuff.
I've also had problems installing CentOS 5.x guests, I don't think I've
managed it yet.
I didn't report it because I don't like "it doesn't work" reports, and
really I had no more than that.
I would feel more confident with that patch in, I'm not sure I want to
go through another rawhide/alpha experience.
The host CPU is Intel E6300 dual core, and I'm running 64-bit host, and
I do have frequency scaling.
1aaaaaaa at coco.merseine.nu Z1aaaaaaa at coco.merseine.nu
You cannot reply off-list:-)
More information about the test