Skip to contentSkip to conversation

kvm-63: Windows Server 2003 randomly hangs with 100% CPU

DISCUSSION Felix Leimbach 1 participant 3 messages Mar 28, 2008 – Apr 1, 2008 KVM

lore
Share & tools
lore
#0 lore raw

Hello list,

I encountered two Windows Server 2003 Std SP1 guests which would randomly hang with 100% CPU after several hours of normal operation. The guest would not respond at all, not even to ping requests. Its process on the host uses 100% CPU.

The Host:
Dual-Core AMD Opteron(tm) Processor 2212
10GB DDRII RAM
Linux 2.6.24.3
KVM-63 userspace
Other KVM guests (Linux and Windows Vista) are rock-solid

The two problematic guests:
Imaged from VMWare ESX 2.5 with reinstall of the ACPI Uniprocessor HAL and drivers.
Switching to the Standard PC HAL and using -no-acpi does not help. Tried with the new virtio NIC first and changed to ne2k_pci later but no effect.
One guest uses a raw hard disk image the other qcow2.

The guests are launched with this command-line:
kvm -hda bonus-system.qcow2 -m 1024 -net nic,model=ne2k_pci,macaddr=52:54:00:74:A1:0C -net tap,ifname=vm-bonus -vnc 10.73.250.1:2 -k de -monitor tcp:10.73.250.1:5952,server,nowait -localtime

Currently I'm testing with a fresh install of Win2003 STD SP1 to rule out the imaged-from-ESX factor.
Also I'm running one of the machines with -no-kvm now and will report back how that works out.

Any idea how to go about debugging this?

Also I see interesting syslog entries. But they don't seem directly related to the hangs, as the times do not match.

Mar 27 14:30:26 kernsrc@obelix kvm: 1363: cpu0 kvm_set_msr_common: 
MSR_IA32_MC0_STATUS 0x0, nop
Mar 27 15:46:52 kernsrc@obelix Ignoring de-assert INIT to vcpu 0
Mar 27 17:20:45 kernsrc@obelix Ignoring de-assert INIT to vcpu 0
Mar 27 17:37:30 kernsrc@obelix apic write: bad size=1 fee00030
Mar 27 17:37:30 kernsrc@obelix Ignoring de-assert INIT to vcpu 0
Mar 27 17:42:19 kernsrc@obelix apic write: bad size=1 fee00030
Mar 27 17:42:19 kernsrc@obelix Ignoring de-assert INIT to vcpu 0
Mar 27 17:55:34 kernsrc@obelix apic write: bad size=1 fee00030
Mar 27 17:55:34 kernsrc@obelix Ignoring de-assert INIT to vcpu 0
Mar 27 20:13:14 kernsrc@obelix apic write: bad size=1 fee00030
Mar 27 20:13:14 kernsrc@obelix Ignoring de-assert INIT to vcpu 0

Cheers,
FL

-------------------------------------------------------------------------
Check out the new SourceForge.net Marketplace.
It's the best place to buy or sell services for
just about anything Open Source.
http://ad.doubleclick.net/clk;164216239;13503038;w?http://sf.net/marketplace

#1 lore raw

Update:
The same mysterious 100% CPU Windows hangs occur with KVM-64 on a 2.6.24.3 kernel.
While I could not yet reproduce the hang on a cleanly installed Windows Server 2003 STD guest, the same hang did occur on an idle Windows Vista guest which was cleanly installed in the KVM environment.
So I guess my transferring the guests from an ESX 2.5 machine does *not* cause the hangs.
The linux guests (2.6.24 kernels) run 100% stable.

I'm still willing to assist in debugging this if a KVM developer is interested.
A TCP monitor port is open and I'm able to run gdb on the hanged machine (if that is of any use with closed-source guests).

Next I'm going to test on 2.6.25-rc7-git5 with KVM-64.

On a sidenote: The help-texts in the 2.6.25 kernel for the new virtio {pci,balloon} drivers do not state, whether the drivers are intended for the guest or the host or on both. I think that would be worth pointing out, like it is done in the CONFIG_KVM help text (the keyword being "hosting" there).

Quoted 53 lines · Felix Leimbach

Hello list,

I encountered two Windows Server 2003 Std SP1 guests which would randomly hang with 100% CPU after several hours of normal operation. The guest would not respond at all, not even to ping requests. Its process on the host uses 100% CPU.

The Host:
Dual-Core AMD Opteron(tm) Processor 2212
10GB DDRII RAM
Linux 2.6.24.3
KVM-63 userspace
Other KVM guests (Linux and Windows Vista) are rock-solid

The two problematic guests:
Imaged from VMWare ESX 2.5 with reinstall of the ACPI Uniprocessor HAL and drivers.
Switching to the Standard PC HAL and using -no-acpi does not help. Tried with the new virtio NIC first and changed to ne2k_pci later but no effect.
One guest uses a raw hard disk image the other qcow2.

The guests are launched with this command-line:
kvm -hda bonus-system.qcow2 -m 1024 -net nic,model=ne2k_pci,macaddr=52:54:00:74:A1:0C -net tap,ifname=vm-bonus

(quoted diff) +0 -2
-vnc 10.73.250.1:2 -k de -monitor tcp:10.73.250.1:5952,server,nowait 
-localtime

Currently I'm testing with a fresh install of Win2003 STD SP1 to rule out the imaged-from-ESX factor.
Also I'm running one of the machines with -no-kvm now and will report back how that works out.

Any idea how to go about debugging this?

Also I see interesting syslog entries. But they don't seem directly related to the hangs, as the times do not match.

Mar 27 14:30:26 kernsrc@obelix kvm: 1363: cpu0 kvm_set_msr_common: 
MSR_IA32_MC0_STATUS 0x0, nop
Mar 27 15:46:52 kernsrc@obelix Ignoring de-assert INIT to vcpu 0
Mar 27 17:20:45 kernsrc@obelix Ignoring de-assert INIT to vcpu 0
Mar 27 17:37:30 kernsrc@obelix apic write: bad size=1 fee00030
Mar 27 17:37:30 kernsrc@obelix Ignoring de-assert INIT to vcpu 0
Mar 27 17:42:19 kernsrc@obelix apic write: bad size=1 fee00030
Mar 27 17:42:19 kernsrc@obelix Ignoring de-assert INIT to vcpu 0
Mar 27 17:55:34 kernsrc@obelix apic write: bad size=1 fee00030
Mar 27 17:55:34 kernsrc@obelix Ignoring de-assert INIT to vcpu 0
Mar 27 20:13:14 kernsrc@obelix apic write: bad size=1 fee00030
Mar 27 20:13:14 kernsrc@obelix Ignoring de-assert INIT to vcpu 0

Cheers,
FL

Quoted 10 lines

-------------------------------------------------------------------------
Check out the new SourceForge.net Marketplace.
It's the best place to buy or sell services for
just about anything Open Source.
http://ad.doubleclick.net/clk;164216239;13503038;w?http://sf.net/marketplace
_______________________________________________
kvm-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/kvm-devel

-------------------------------------------------------------------------
Check out the new SourceForge.net Marketplace.
It's the best place to buy or sell services for
just about anything Open Source.
http://ad.doubleclick.net/clk;164216239;13503038;w?http://sf.net/marketplace

#2 lore raw
Quoted 1 line

Next I'm going to test on 2.6.25-rc7-git5 with KVM-64.

Running on 2.6.25-rc7-git5 with KVM 64 for 48 hours now and no more hangs!
Note that the longest uptime of the Windows 2003 Std guest was around 20 hours before.
So it *seems* that 2.6.25 does indeed solve the problem, although a couple more days of testing is needed to be sure.

-------------------------------------------------------------------------
Check out the new SourceForge.net Marketplace.
It's the best place to buy or sell services for
just about anything Open Source.
http://ad.doubleclick.net/clk;164216239;13503038;w?http://sf.net/marketplace

_______________________________________________
kvm-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/kvm-devel