Re: Running KVM inside a chroot

[Avi Kivity <avi@redhat.com>]

On 10/13/2011 02:51 AM, Jorge Lucangeli Obes wrote:
> On Wed, Oct 12, 2011 at 3:52 PM, Jorge Lucangeli Obes
> <jorgelo@chromium.org> wrote:
> > On Wed, Oct 12, 2011 at 12:55 PM, Alexander Graf <agraf@suse.de> wrote:
> >>
> >> On 12.10.2011, at 20:49, Jorge Lucangeli Obes wrote:
> >>
> >>> Hi all,
> >>>
> >>> I'm working on Chromium OS development. We have a pretty elaborate
> >>> chroot inside of which we carry out all development. We use KVM to
> >>> launch Chromium OS builds inside a VM for testing. Turns out that for
> >>> some reason, when QEMU is launched from inside the chroot, KVM itself
> >>> seems not to be used. The VM is extremely slow.
> >>>
> >>> Is this known/expected? QEMU is installed inside the chroot, the KVM
> >>> modules are loaded, the /dev/kvm device is present and accesible. Any
> >>> ideas on how to debug this?
> >>
> >> The first obvious idea I'd have here would be to strace the qemu process and check what happens when it opens /dev/kvm :)
>
> Resending since original attachment was too large.
>
> > That's what I thought. I did a test run under strace. I'm attaching
> > the list of syscalls from the call to 'open(/dev/kvm)' to the first
> > successful 'ioctl(KVM_RUN)'. /dev/kvm seems to be opened correctly, a
> > VCPU is created, and then that VCPU is used with KVM_RUN. After the
> > first call to 'ioctl(KVM_RUN)', there are long lists of more KVM_RUN
> > calls, separated by brief groups of other calls. So, IIUC, KVM seems
> > to be used, and seems to be "working", but the VM is one order of
> > magnitude slower anyways.
> >
> > Any ideas?
>

What do top/vmstat/kvm_stat say?

-- 
error compiling committee.c: too many arguments to function