Re: [Qemu-devel] [PATCH 1/2] virtio-9p: print error message and exit instead of BUG_ON()

[Greg Kurz <groug@kaod.org>]

On Fri, 9 Sep 2016 10:53:05 +0200
Cornelia Huck <cornelia.huck@de.ibm.com> wrote:

> On Fri, 9 Sep 2016 10:46:25 +0200
> Greg Kurz <groug@kaod.org> wrote:
> 
> > On Fri, 9 Sep 2016 10:30:53 +0200
> > Cornelia Huck <cornelia.huck@de.ibm.com> wrote:
> >   
> > > On Thu, 8 Sep 2016 19:55:16 +0300
> > > "Michael S. Tsirkin" <mst@redhat.com> wrote:
> > >   
> > > > On Thu, Sep 08, 2016 at 06:26:52PM +0200, Greg Kurz wrote:    
> > > > > On Thu, 8 Sep 2016 18:19:27 +0300
> > > > > "Michael S. Tsirkin" <mst@redhat.com> wrote:
> > > > >     
> > > > > > On Thu, Sep 08, 2016 at 05:04:47PM +0200, Cornelia Huck wrote:    
> > > > > > > On Thu, 8 Sep 2016 18:00:28 +0300
> > > > > > > "Michael S. Tsirkin" <mst@redhat.com> wrote:
> > > > > > >       
> > > > > > > > On Thu, Sep 08, 2016 at 11:12:16AM +0200, Greg Kurz wrote:      
> > >   
> > > > > If it continues
> > > > > execution, this means we're expecting the guest or the host to do something
> > > > > to fix the error condition. This requires QEMU to emit an event of some
> > > > > sort, but not necessarily to log an error message in a file. I guess this
> > > > > depends if QEMU is run by some tooling, or by a human.    
> > > > 
> > > > I'm not sure we need an event if tools are not expected to
> > > > do anything with it. If we limit # of times error
> > > > is printed, tools will need to reset this counter,
> > > > so we will need an event on overflow.    
> > > 
> > > If the device goes into a broken state, it should be discoverable from
> > > outside. I'm not sure we need an actual event signalling this if this
> > > happens due to the guest doing something wrong: That would be a task
> > > for tools monitoring _inside_ the guest.   
> > 
> > Well, in case of a virtio device being broken, section 2.1.2 in the spec
> > suggests to set the status to DEVICE_NEEDS_RESET and to notify it to
> > the guest (aka. event signalling). I'll send a patch shortly.  
> 
> Stefan had already sent
> <1460467534-29147-4-git-send-email-stefanha@redhat.com> ages ago, but
> it has not yet made it anywhere...
> 

I don't know what to do with this message-id :\

> Anyhow, I was concerned with host signalling (sorry for being unclear),
> and I still do not think we need to alert host monitoring software to
> guest stupidity.
> 

I agree. Sorry if my poor wording made you (and others) think I was
suggesting that :) My point was that if QEMU exits because of guest
stupidity, you are forced to error_report() something to the host,
but this is really suboptimal (even if BUG_ON is worse)... then
there was that discussion about log files getting to big, but I don't
even know how we came there, as it does not really make sense when QEMU
exits.

> >   
> > > For tools monitoring the
> > > health of the machine (from the host perspective), the discovery
> > > interface would probably be enough?
> > >   
> > 
> > Yeah, probably.
> > 
> > Cheers.
> > 
> > --
> > Greg
> >   
>