Re: [Xen-devel] [PATCH v4] libxl: add basic spice support for pv domUs

[Fabio Fantoni <fabio.fantoni@xxxxxxx>]

Il 09/05/2014 16:00, Ian Campbell ha scritto:
> On Fri, 2014-05-09 at 15:48 +0200, Fabio Fantoni wrote:
>
> > > > +    libxl_defbool_setdefault(&b_info->u.hvm.spice.enable, false);
> > > > +    if (libxl_defbool_val(b_info->u.hvm.spice.enable)) {
> > > I think this will cause hvm.spice to take precedence over the top-level
> > > spice field, which I don't think is what you wanted, was it?
> > >
> > > I think you need:
> > >       libxl_defbool_setdefault(&b_info->u.hvm.spice.enable, false);
> > >       libxl_defbool_setdefault(&b_info->spice.enable, false);
> > >       if (!libxl_defbool_val(&b_info->spice.enable) &&
> > >           libxl_defbool_setdefault(&b_info->u.hvm.spice.enable))
> > >               b_info->spice = b_info->u.hvm.spice;
> > >
> > > Then this bit:
> > >
> > > > +    if (libxl_defbool_val(b_info->spice.enable)) {
> > > > +        libxl_defbool_setdefault(&b_info->spice.disable_ticketing, false);
> > > > +        libxl_defbool_setdefault(&b_info->spice.agent_mouse, true);
> > > Does this work for PV guests? Having the default vary with b_info->type
> > > would be fine IMHO.
> > >
> > > I don't see you changing the result of calling libxl__need_xenpv_qemu
> > > anywhere to be true if spice is enabled. I think this means qemu won't
> > > be launched unless there happens to also be a qdisk or some other qemu
> > > backend.
> > Spice requires qemu.
> Understood.
>
> > What would I do to always start qemu with spice enabled? (even if unused
> > qdisk and vnc and/or using api instead of xl)
> You need to update libxl__need_xenpv_qemu to return true if spice is
> enabled, along with all the other conditions.
Now I understand, thanks.

> But: Does spice already require a VFB? In which case the existing
> handing of that will suffice.
>
> I think I'm probably confused about something: Does this patch
> enable/expose SPICE to the guest? Or does it simply enable the existing
> PVFB's output to be exposed from the qemu process to a spice client? I
> had been thinking this was a parallel feature to PVFB, but I'm starting
> to suspect this might actually be an additional feature *of* PVFB, which
> is correct? (I'm afraid that depending on which it is I might have to
> check back over my review to make sure I haven't suggested anything
> stupid)
>
> Does QXL change that answer?
QXL and other emulated vga is now not supported on pv domUs because 
require an emulated pci bus and MMIO not supported on pv domUs FWIK.
During my first test, I was enabled spice in pv domus but the guest wont 
show nothing while connecting to it.
On my second shot, I've added -vga xenfb but it is deprecated. This way 
I've got at least video out but no mouse nor keyboard.
Following the suggestions from Stefano Stabellini I've then activated 
vfb the same way as for vnc and i got it all working but mouse, which is 
only visible in some circumstances (such grafical debian installer) but 
always working (even if not visible).
The last one is the actual patch.
Probably the vfb part is not complete or at least improvable.
Perhaps it is even not necessary if pvfb will became modified to works 
with spice, so domUs use this through its modules.
I feel I'm not perfectly undrestanding pvfb. Please would you like 
better describe me this component?
Another thing: i noticed that vnc has all its parameters duplicated in 
vfb too. For the moment I omissed them since i feel they are not 
necessary - aren't they?
Thanks so much for any replies.

> > > Does spice not support multiple interfaces like the vfb option does? If
> > > so we should plan for that in the interface I think.
> > I not know for sure what spice multiple interfaces support is.
> I think the importance of this depends on my question about VFB.
>
> Ian.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel