After installing Windows 2000 guest with sp4 all seems work fine. But after short time Windows download a large pack of updates (about 68), when I installed this updates and rebooted guest host CPU utilization rised to 100% even if guest is idle. Before installing updates this problem was not observed. I trying with two Win2k installation with the same results. Host CPU utilization is equal to number of vCPUs, for example with -smp 2 host CPU usage is 200%.
Host:
2x Quad Core Opteron 8356
64GB RAM
OS: Gentoo 64bit
kernel: 2.6.25
KVM-78 (modules from package not from kernel)
Guest:
Windows 2000 sp4 32bit started with command:
kvm -M pc -m 2048 -smp 2 -name Win200032sp4 -monitor pty -localtime -boot d -drive file=/dev/MDvg_InstallVM/Win200032sp4,if=ide,index=0 -drive file=/mnt/MD_isos/Windows32/pl_win2000_pro_win2000_server.iso,if=ide,media=cdrom,index=2 -net nic,macaddr=00:16:3e:44:b2:45,vlan=0,model=virtio -net tap,fd=44,script=,vlan=0,ifname=vnet9
On the guest was installed virtio net driver.
On the same host I running several other guests systems including Win2003 Server sp2 32bit, WinXP sp2 32bit and Linux Gentoo 64bit without problem like this.
Using -no-acpi switch not gives any difference.
I can confirm this bug. On my setup it only apperas when Windows is setup to use more than one core (Device Manager -> Computer -> MPS multi processor...). If I go to "driver" -> "update driver" and install the "MPS single core" driver, the guest runs at normal cpu usage, but uses one core only (even with kvm's "-smp n" command line option).
Windows 2000 Server (SP4) is also affected by this bug. Exactly the same behaviour.
My System:
Intel Core 2 Quad Q9550 (with intel vt support)
8G of DDR2-RAM
OS: Debian Sid
kernel: 2.6.27.7 (last stable from kernel.org)
KVM-79
Guests:
Windows 2000 Professional SP4 (affected by this bug)
Windows 2000 Server SP4 (also affected by this bug)
Debian lenny (not affected)
pfSense 1.2.1 (not affected)
Executed command:
qemu-system-x86_64 -smp 4 -no-acpi -hda win2k-server.qcow -drive file=../iso-images/win2k-server.iso,if=ide,media=cdrom -boot c -m 2048 -net nic,macaddr=DE:AD:BE:EF:12:19 -net tap,script=/etc/kvm/kvm-ifup-br0 -daemonize
This results in a process that shows up with a CPU usage of about 400% in "top".
Upgrading to KVM-79 doesn't help.
I am also seeing this behaviour with kvm-81. Guest is idle according to Windows task manager, but the kvm process on the host is using 200% CPU.
Host:
Intel Core2 Duo E8400
4GB RAM
Debian Sid/Unstable
Kernel 2.6.26-1-amd64
kvm-81 (compiled from upstream source, not Debian packages)
Guest:
Windows 2000 SP4
/usr/local/kvm/bin/qemu-system-x86_64 \
-smp 2 \
-localtime -m 1024 \
-hda disk1.img \
-net nic,vlan=0,macaddr=52:54:00:12:A0:01,model=rtl8139 \
-net tap,vlan=0,ifname=tap0,script=no \
-vnc 127.0.0.1:1 -usbdevice tablet \
-daemonize
I have the same problem, but seems this is win2k 's problem. I watch a post saying that ACPI work fine. Seems it is the idle loop that do not issue hlt because it is not acpi and I did try it and it work for me.
In the guest, enable ACPI and do a new installation, it will become "ACPI Multiprocessor" and the host would not have high cpu usage.
For some reason, it won't work and do not have the options of "ACPI Multiprocessor" if u already do a "Standard PC" installation. You need some tweak to change it.
new installation with ACPI do not have this problem. Be sure that "ACPI Multiprocessor" is choose instead of "Standard PC"
My installation definitely shows "ACPI Multiprocessor PC" in the device manager (I don't think I could have SMP without it).
roy-anonymous: can you give some more details about your host environment? cpu type, kernel, kvm version, kvm command line...
By the way, I may as well confirm this is still a problem for me with a 2.6.28 kernel and kvm-82.
Sorry for my late reply.
Here is my configuration.
Host:
Intel Q9550
8GB RAM
OS: Fedora 10 64bit
Kernel: 2.6.27.9-159.fc10.x86_64
KVM: kvm-74-10.fc10.x86_64 (All come with yum)
Guest:
Windows 2000 sp4 32bit:
Start with virt-manager that come with Fedora 10.
Yes, you could have SMP without ACPI one.
For me, If I do a new installation with ACPI Multiprocessor (with acpi enabled in libvirt), it's working fine.(also try tested with virtio net driver installed)
If I use "Multiprocessor PC" (with acpi disabled) it will have high cpu usage problem.
May be this is really the problem of different distro or the new versions of KVM?
Roy, you didn't say whether you have fully patched your install with all the post-sp4 updates. As the original reported said, the host CPU usage problem doesn't occur until after the post-sp4 updates have been applied.
Just confirming that the problem is still present with kvm-84 (on Linux 2.6.28.7).
I spent a whole lot of time today trying to work out which patch causes the host CPU usage to peg at 100% (even though the guest task manager shows everything is idle). As the original poster says, a clean install of Windows 2000 and up to service pack 4 works fine.
As soon as I installed "Update Rollup 1 for Windows 2000 SP4 (KB891861)", the host CPU usage problem starts. I can uninstall the rollup package and the problem goes away again.
http://support.microsoft.com/?kbid=891861
I tried going through all the individual updates that are supposed to be included in the rollup (listed on the KB page above), but installing them individually never made the issue come up. So it must be something else that is included in the rollup package but not documented I guess... something which modifies the method the OS uses for idling? FWIW, whatever is causing the host CPU usage to go so high does not cause the CPUs to get hot. They remain cool as if they were idle.
This test was done on a dual Intel Xeon 5130 machine with 32GB RAM. Linux kernel 2.6.30, kvm-86.
/usr/local/kvm/bin/qemu-system-x86_64 \
-smp 2 \
-localtime -m 2048 \
-drive if=ide,index=0,media=disk,file=disk1.img,boot=on \
-drive if=ide,index=1,media=cdrom \
-drive if=ide,index=2,media=disk,file=disk2.img \
-net nic,vlan=0,macaddr=52:54:00:12:34:6A,model=virtio \
-net tap,vlan=0,ifname=tap20,script=no \
-vnc 127.0.0.1:20 \
-usbdevice tablet \
-daemonize
Hi,
Do you still see this problem? If so, would you mind opening a bug about it over at launchpad.net?
Thanks,
Jes