Re: [Qemu-devel] [PATCH v2] hmp/qxl: info spice: add qxl info

[Alon Levy]

On Tue, May 29, 2012 at 01:44:35PM -0300, Luiz Capitulino wrote:
> On Tue, 29 May 2012 17:51:50 +0300
> Alon Levy <address@hidden> wrote:
> 
> > On Tue, May 29, 2012 at 10:38:20AM -0300, Luiz Capitulino wrote:
> > > On Tue, 29 May 2012 09:25:40 +0200
> > > Gerd Hoffmann <address@hidden> wrote:
> > > 
> > > >   Hi,
> > > > 
> > > > >> How would that work? I have QXLInfo that only makes sense when the
> > > > >> information is about a qxl device. Can't have opaque data in a QMP
> > > > >> response, so would this be a "info display qxl" "info display cirrus"
> > > > >> etc. or "info qxl"?
> > > > > 
> > > > > You could show what's available and/or being used currently.
> > > > 
> > > > I think (Alon, correct me if I'm wrong) the use case is being able to
> > > > figure whenever the guest drivers are installed and active.
> > > 
> > > Alon, can you confirm this? I'm still not clear on the use-case.
> > > 
> > > The two points I'm wondering are 1. If this is really tied to spice (and 
> > > thus
> > > this info should be part of query-spice) and now 2. if the use case above
> > > really pertains to QMP.
> > > 
> > > I've talked to someone in the past about having a way to get information 
> > > about
> > > guest drivers statuses and the conclusion was that the guest-agent agent 
> > > was
> > > better suited for that.
> > 
> > The information I'm suggesting to expose is state of the guest as seen
> > from device pov:
> 
> Do you expect mngt apps to use that info or is it just for debugging? If it's
> the latter, then maybe we could add query-qxl as gen=no and declare it 
> unstable.

It's for debugging. I don't expect management to use it, but I do expect
it to be useful when someone comes with a problem to the list, as it's
easy to retrieve information.

Where do I add "gen=no", how do you want me to mark this unstable? The
point is to have this available, not to have this disabled. It's low
risk, it's only displayed when someone does "info spice" (or "info virt"
if someone would say that is preferable) or the equivalent QMP command.

> 
> >  * guest_bug - this indicates that the qxl device has been instructed to
> >   do something illegal that it has ignored, and until a reset from the
> >   guest, (QXL_RESET io write), that would presumably indicate a restart
> >   of the guest driver, we would like to ignore the guest from now on.
> >   This information would be helpful for error report.
> >
> >  * mode - this is another indication of presence or lack of a guest
> >   driver. This time more straight forward - if a certain IO is issued
> >   (QXL_CREATE_PRIMARY) then the driver is in native mode. If another IO
> >   (SET_MODE) we are in compat mode. If anything to change back to vga
> >   happens (vga io read/write), we are back in vga, and if a
> >   DESTROY_SURFACE happens we are in undefined mode. This status would be
> >   helpful for error reporting as well.
> 
> What kind of error report?

If someone reports an error involving spice/qxl this value would be a
quick indication of false alarm for some instances, i.e. showing vga
would probably mean there was no driver loaded in the guest.

> 
> > 
> > Since both pieces of information already exist in the qemu side, there
> > is no need to introduce an agent to retrieve them. They are easy to
> > retrive with existing management (libvirt/vdsm can do random monitor
> > commands iirc).
> > 
> > Alon
> > 
>