Re: [Ilw] iwlagn and kvm related "BUG: scheduling while atomic" after resuming

[Avi Kivity <avi@redhat.com>]

On 07/19/2011 05:23 PM, Stefan Hajnoczi wrote:
> >
> >  Can you post some disassembly around %rip?
>
> Here is the kvm.ko disassembly for %rip = kvm_arch_vcpu_ioctl_run+0x927.
>
> I think at runtime the pvops bit should be patched and we should just
> have a sti instruction?

Yes, it's a local_irq_enable() (followed by a 4-byte nop; %rip points 
past the nop due to interrupt shadow).

> I don't understand the backtrace.  It prints out the interrupt handler
> stack frames but %rip appears to be down in kvm.ko.  If there is an
> interrupt handler running shouldn't %rip be in there?

The interrupt handler finished running; just before returning it called 
the scheduler (which it shouldn't have, since we're in a 
preempt_disabled() section).

> Unfortunately the full oops scrolled off the screen so I only have
> panic1.jpg and panic2.jpg.  We're missing earlier information.  Is it
> possible that there was another panic first and that kvm.ko is just
> dying in the aftermath?

It's possible.  Is this reproducible?  If so please set up netconsole or 
kdump.

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