USB tablet CPU usage pattern. Possible bug?

USB tablet CPU usage pattern. Possible bug?

[Thomas Løcke]

Hey all,

The other day I upgraded the kernel on one of my KVM hosts. I went
from 2.6.34.1 to 2.6.35.7, and immediately I noticed that my Windows
XP guests was now using significantly more CPU while idle, compared to
the 2.6.34.1 kernel. All the Windows XP guests are running with
-usbdevice tablet.

Using the 2.6.34.1 kernel idle CPU usage for the Windows XP guests was
sitting at ~5%, with spikes going as high as 10%. Using 2.6.35.7 these
numbers were ~20%, with spikes going as high as 35%.

Everything appeared to work as usual, except for this higher idle load.

I'm using qemu-kvm 0.12.50. All images are raw. No SMP for the Windows
XP guests.

I downgraded to 2.6.34.7, and the CPU load pattern is now back to normal.

The server is a Sun Fire X4270 (dual quad core Xeon 5520, 24GB RAM)
running Slackware 13.1 x86_64.

I've no idea whether this is intentional, or if it is a bug in the
kernel, the KVM modules or in qemu-kvm.

Regards,
Thomas Løcke

Re: USB tablet CPU usage pattern. Possible bug?

[Michael Tokarev]

05.10.2010 09:36, Thomas Løcke wrote:
> Hey all,
> 
> The other day I upgraded the kernel on one of my KVM hosts. I went
> from 2.6.34.1 to 2.6.35.7, and immediately I noticed that my Windows
> XP guests was now using significantly more CPU while idle, compared to
> the 2.6.34.1 kernel. All the Windows XP guests are running with
> -usbdevice tablet.
> 
> Using the 2.6.34.1 kernel idle CPU usage for the Windows XP guests was
> sitting at ~5%, with spikes going as high as 10%. Using 2.6.35.7 these
> numbers were ~20%, with spikes going as high as 35%.
> 
> Everything appeared to work as usual, except for this higher idle load.
> 
> I'm using qemu-kvm 0.12.50. All images are raw. No SMP for the Windows
> XP guests.
> 
> I downgraded to 2.6.34.7, and the CPU load pattern is now back to normal.
> 
> The server is a Sun Fire X4270 (dual quad core Xeon 5520, 24GB RAM)
> running Slackware 13.1 x86_64.
> 
> I've no idea whether this is intentional, or if it is a bug in the
> kernel, the KVM modules or in qemu-kvm.

It's not only usb tablet, it's any usb device.  There were a long
thread started by me a while back, initially titled

 high load with win7 & usb tablet

and later renamed to

 high load with usb device

here on kvm@vger.  But there's nothing conclusive in there.

There are also several threads on LKML (linux kernel mailing
list) about high power consumption and high idle load with
recent (2.6.35+ and even some stable) kernels, also without
anything conclusive.

The two (high cpu utilisation in kvm while guest is only
checking usb devices and high idle load in recent kernels)
may be related or may be not.  It'd be nice to find out
what the problem(s) is(are), but so far it weren't done.

/mjt