Re: [Qemu-trivial] [PATCH for 2.1] vfio: use correct runstate

[Paolo Bonzini]

----- Messaggio originale -----
> Da: "Alex Williamson" <address@hidden>
> A: "Paolo Bonzini" <address@hidden>
> Cc: address@hidden, address@hidden
> Inviato: Venerdì, 27 giugno 2014 18:34:59
> Oggetto: Re: [PATCH for 2.1] vfio: use correct runstate
> 
> On Fri, 2014-06-27 at 16:32 +0200, Paolo Bonzini wrote:
> > io-error is for block device errors; it should always be preceded
> > by a BLOCK_IO_ERROR event.
> 
> Where does this requirement come from?  I only see a loose association
> of IO_ERROR to disk in libvirt and none in QEMU.

See the RunState enum in qapi-schema.json:

##
# @RunState
#
# An enumeration of VM run states.
#
# ...
#
# @internal-error: An internal error that prevents further guest execution
# has occurred
#
# @io-error: the last IOP has failed and the device is configured to pause
# on I/O errors
#
# @paused: guest has been paused via the 'stop' command

The point of io-error is that management can look at block devices, see if
any have an error reported, and then resume execution (see documentation of
rerror=stop and werror=stop/enospc).  This is counter to the intentions you
have in vfio.

> > I think vfio wants to use
> > RUN_STATE_INTERNAL_ERROR instead.
> 
> But that seems to put us into an "unknown" paused state in libvirt.

I think paused is incorrect, because (unlike RUN_STATE_IO_ERROR), you cannot
resume from RUN_STATE_INTERNAL_ERROR except with a reset.  QEMU enforces that,
and this matches the error you are reporting:

    error_report("%s(%04x:%02x:%02x.%x) Unrecoverable error detected.  "
                 "Please collect any data possible and then kill the guest",
                 __func__, vdev->host.domain, vdev->host.bus,
                 vdev->host.slot, vdev->host.function);

libvirt has a crashed state, I think that's what libvirt should call the
internal-error runstate.  IIRC on Xen you get to crashed when the processor
raises an error on vmentry, for example.

Libvirt only knows about crashed/unknown, but one could add 
crashed/internal-error
too.

Paolo