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

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

[Fabio Fantoni]

Il 13/05/2014 11:50, Ian Campbell ha scritto:
> On Tue, 2014-05-13 at 10:00 +0200, Fabio Fantoni wrote:
>>> 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.
> Right, but what about the previous paragraph? That was the important bit
> for me to understand, since it impacts the entire configuration model
> and libxl API.
>
>> 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.
> Is SPICE a mechanism for exposing the PVFB or is it something entirely
> separate? That is the crux of my question above.

I added also qemu-devel and spice-devel on cc:
Any one on this please?

>
>> I feel I'm not perfectly undrestanding pvfb. Please would you like
>> better describe me this component?
> I need to understand what it is this patch is actually doing. I'm
> starting to worry that perhaps you don't understand either.

Make possible using spice with basic features also on pv domUs.
Since emulated vgas is not supported on pv I use vfb instead, even if 
this part is incomplete and rudimental.
I'm waiting from some aswers from spice and qemu teams to share some 
light on the question above.

Thanks for any reply and sorry for my bad english.

>
>> 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
> It really depends on the answer to my questions above.
>
> If SPICE is nothing to do with PVFB then it obviously makes no sense for
> it to be separate.
>
> If SPICE is just a backend for PVFB then it actually makes *more* sense
> to expose spice as a PVFB configuration parameter than as a set of top
> level options. The fact that VNC is exposed as a top level thing too is
> somewhat anomalous but was done for xend compatibility (nb: it is the
> top level duplicating the vfb VNC settings, not vice versa IMHO).
>
> Ian.
>

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

[Ian Campbell]

On Tue, 2014-05-13 at 12:51 +0200, Fabio Fantoni wrote:
> Il 13/05/2014 11:50, Ian Campbell ha scritto:
> > On Tue, 2014-05-13 at 10:00 +0200, Fabio Fantoni wrote:
> >>> 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.
> > Right, but what about the previous paragraph? That was the important bit
> > for me to understand, since it impacts the entire configuration model
> > and libxl API.
> >
> >> 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.
> > Is SPICE a mechanism for exposing the PVFB or is it something entirely
> > separate? That is the crux of my question above.
> 
> I added also qemu-devel and spice-devel on cc:
> Any one on this please?
> 
> >
> >> I feel I'm not perfectly undrestanding pvfb. Please would you like
> >> better describe me this component?
> > I need to understand what it is this patch is actually doing. I'm
> > starting to worry that perhaps you don't understand either.
> 
> Make possible using spice with basic features also on pv domUs.

*How* is the important thing. How is SPICE exposed to the guest? How is
SPICE exposed to the user/client?

I'm a bit surprised you've been pushing a patch to enable something
without knowing the answers to these sorts questions. If you don't
understand what it does how do you know it is doing the correct thing,
and how can you therefore explain it to the person reviewing the patch?

> Since emulated vgas is not supported on pv I use vfb instead, even if 
> this part is incomplete and rudimental.

What is incomplete and rudimentary?

> I'm waiting from some aswers from spice and qemu teams to share some 
> light on the question above.
> 
> Thanks for any reply and sorry for my bad english.
> 
> >
> >> 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
> > It really depends on the answer to my questions above.
> >
> > If SPICE is nothing to do with PVFB then it obviously makes no sense for
> > it to be separate.
> >
> > If SPICE is just a backend for PVFB then it actually makes *more* sense
> > to expose spice as a PVFB configuration parameter than as a set of top
> > level options. The fact that VNC is exposed as a top level thing too is
> > somewhat anomalous but was done for xend compatibility (nb: it is the
> > top level duplicating the vfb VNC settings, not vice versa IMHO).
> >
> > Ian.
> >

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

[Fabio Fantoni]

Il 13/05/2014 13:00, Ian Campbell ha scritto:
> On Tue, 2014-05-13 at 12:51 +0200, Fabio Fantoni wrote:
>> Il 13/05/2014 11:50, Ian Campbell ha scritto:
>>> On Tue, 2014-05-13 at 10:00 +0200, Fabio Fantoni wrote:
>>>>> 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.
>>> Right, but what about the previous paragraph? That was the important bit
>>> for me to understand, since it impacts the entire configuration model
>>> and libxl API.
>>>
>>>> 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.
>>> Is SPICE a mechanism for exposing the PVFB or is it something entirely
>>> separate? That is the crux of my question above.
>> I added also qemu-devel and spice-devel on cc:
>> Any one on this please?
>>
>>>> I feel I'm not perfectly undrestanding pvfb. Please would you like
>>>> better describe me this component?
>>> I need to understand what it is this patch is actually doing. I'm
>>> starting to worry that perhaps you don't understand either.
>> Make possible using spice with basic features also on pv domUs.
> *How* is the important thing. How is SPICE exposed to the guest?

FWIK spice-server is inside qemu, used but not "visible" inside the 
guest (seems similar to qemu's vnc in hvm domUs).
Inside the guest there are some advanced and optional features visible 
(installing software and additional drivers like spice-guest-tools).
The advanced features are actually not supported in the pv domUs for 
some missed requirements (for example virtio-serial and emulated pci).

> How is
> SPICE exposed to the user/client?

Users can use these clients to connect to guest using spice: 
remote-viewer (mainly used), spicy or spicec (deprecated).
These clients basically connect to domU's qemu using spice and also to 
some additional software/drivers installed in domU's S.O. for some 
advanced features (if enabled).

>
> I'm a bit surprised you've been pushing a patch to enable something
> without knowing the answers to these sorts questions. If you don't
> understand what it does how do you know it is doing the correct thing,
> and how can you therefore explain it to the person reviewing the patch?

Unfortunately at the moment seems there is no other person available to 
implementing full spice support into xen, so I'm trying to do it myself 
even if with a very short knowledge about it and few time.
Spice is a lot better then what I have used since now (xen's vnc + guest 
rdp and proprietary usb redirection).
My main goal is to use spice full featured on a thin client system to 
gain pc-like experience accessing hvm and pv domUs.
Also use different architectures (for example starting with arm on thin 
client) for now impossibile with proprietary usbredirection (only for 
x86) and with some problem that usbredirect used by spice hasn't ecc...

Spice requires the same qemu parameters on both pv and hvm domUs.
The main difference is that with pv which actually lack of video out and 
keyboard/pointer input, I have used the only thing I found working: vfb 
(+vkb).

>
>> Since emulated vgas is not supported on pv I use vfb instead, even if
>> this part is incomplete and rudimental.
> What is incomplete and rudimentary?

The patch's part about setting/use of vfb(+vkb) with spice on pv domUs.

>
>> I'm waiting from some aswers from spice and qemu teams to share some
>> light on the question above.
>>
>> Thanks for any reply and sorry for my bad english.
>>
>>>> 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
>>> It really depends on the answer to my questions above.
>>>
>>> If SPICE is nothing to do with PVFB then it obviously makes no sense for
>>> it to be separate.
>>>
>>> If SPICE is just a backend for PVFB then it actually makes *more* sense
>>> to expose spice as a PVFB configuration parameter than as a set of top
>>> level options. The fact that VNC is exposed as a top level thing too is
>>> somewhat anomalous but was done for xend compatibility (nb: it is the
>>> top level duplicating the vfb VNC settings, not vice versa IMHO).
>>>
>>> Ian.
>>>
>