Re: [Qemu-devel] [RFC PATCH] vl: fix migration when watchdog expires

["Dr. David Alan Gilbert" <dgilbert@redhat.com>]

* Paolo Bonzini (pbonzini@redhat.com) wrote:
> On 14/08/2018 12:48, Jay Zhou wrote:
> > I got the following error when migrating a VM with watchdog
> > device:
> > 
> > {"timestamp": {"seconds": 1533884471, "microseconds": 668099},
> > "event": "WATCHDOG", "data": {"action": "reset"}}
> > {"timestamp": {"seconds": 1533884471, "microseconds": 677658},
> > "event": "RESET", "data": {"guest": true}}
> > {"timestamp": {"seconds": 1533884471, "microseconds": 677874},
> > "event": "STOP"}
> > qemu-system-x86_64: invalid runstate transition: 'prelaunch' -> 'postmigrate'
> > Aborted
> > 
> > The run state transition is RUN_STATE_FINISH_MIGRATE to RUN_STATE_PRELAUNCH,
> > then the migration thread aborted when it tries to set RUN_STATE_POSTMIGRATE.
> > There is a race between the main loop thread and the migration thread I think.
> 
> In that case I think you shouldn't go to POSTMIGRATE at all, because the
> VM has been reset.

Migration has the VM stopped; it's not expecting the state to change at
that point.

> Alternatively, when the watchdog fires in RUN_STATE_FINISH_MIGRATE
> state, it might delay the action until after the "cont" command is
> invoked on the source, but I'm not sure what's the best way to achieve
> that...

Jay: Which watchdog were you using?

 a) Should the watchdog expire when the VM is stopped; I think it
shouldn't - hw/acpi/tco.c uses a virtual timer as does i6300esb; so
is the bug here that the watchdog being used didn't use a virtual
timer?

 b) If the watchdog expires just before the VM gets stopped, is there
a race which could hit this?  Possibly.

 c) Could main_loop_should_exit guard all the 'request's by
something that checks whether the VM is stopped?

Dave


> Paolo
--
Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK