Re: [Qemu-devel] [PATCH v8] kvm: notify host when the guest is panicked

[Marcelo Tosatti]

On Tue, Aug 14, 2012 at 10:47:48AM +0300, Gleb Natapov wrote:
> On Mon, Aug 13, 2012 at 05:24:52PM -0300, Marcelo Tosatti wrote:
> > On Mon, Aug 13, 2012 at 01:48:39PM -0600, Eric Blake wrote:
> > > On 08/13/2012 12:21 PM, Marcelo Tosatti wrote:
> > > > On Wed, Aug 08, 2012 at 10:43:01AM +0800, Wen Congyang wrote:
> > > >> We can know the guest is panicked when the guest runs on xen.
> > > >> But we do not have such feature on kvm.
> > > >>
> > > >> Another purpose of this feature is: management app(for example:
> > > >> libvirt) can do auto dump when the guest is panicked. If management
> > > >> app does not do auto dump, the guest's user can do dump by hand if
> > > >> he sees the guest is panicked.
> > > >>
> > > >> We have three solutions to implement this feature:
> > > >> 1. use vmcall
> > > >> 2. use I/O port
> > > >> 3. use virtio-serial.
> > > >>
> > > >> We have decided to avoid touching hypervisor. The reason why I choose
> > > >> choose the I/O port is:
> > > >> 1. it is easier to implememt
> > > >> 2. it does not depend any virtual device
> > > >> 3. it can work when starting the kernel
> > > > 
> > > > How about searching for the "Kernel panic - not syncing" string 
> > > > in the guests serial output? Say libvirtd could take an action upon
> > > > that?
> > > > 
> > > > Advantages:
> > > > - It works for all architectures.
> > > > - It does not depend on any virtual device.
> > > 
> > > But it _does_ depend on a serial console,
> > 
> > Which already exists and is supported.
> > 
> > >  and furthermore requires
> > > libvirt to tee the serial console (right now, libvirt can treat the
> > > console as an opaque pass-through to the end user, but if you expect
> > > libvirt to parse the serial console for a particular string, you've lost
> > > some efficiency).
> > > 
> > > > - It works as early as serial console output does (panics before
> > > > that should be rare).
> > > > - It allows you to see why the guest panicked.
> > > 
> > > I think your arguments for a serial console have already been made and
> > > refuted in earlier versions of this patch series, which is WHY this
> > > series is still applicable.
> > 
> > Refuted why, exactly? Efficiency to parse serial console output in
> > libvirt should not be a major issue surely?
> > 
> It is not zero config (guests do not send console output to serial by
> default). If vm users want to use serial for its working console panic
> notification will trigger every time user examines dmesg with "Kernel
> panic - not syncing" in it.

Ok, then it would have to be a dedicated serial console which starts
to become funny.

Use a simple virtio device, then, it starts early enough (or can be made
to) during kernel init for most relevant production panics, and works
for all architectures.