libvirt users - May 2019 - Re: [iGVT-g] GVT-g - suboptimal user experience

Hello.

All features are about usability and simple user experience.

1. Its about local display / dmabuf feature. Currently user needs to use
virt-viewer tool. But virt-manager already incorporates graphical console. It
would be nice if it could support accelerated gvt-g local display. Preferably
with minimum performance overhead. Also virt-manager should allow to use mdev
videocard alone, without requiring to use additional primary virtual adapter
like QXL or Cirrus

2. Currently with gvt solution there few mdev devices each with fixed
resolution. User should be able to configure own resolutions for devices at host
without additional manipulations in guest. Ideally, resolution should be dynamic
and change appropriately on resize of virt-manager window

3. This is about windows guest and VM configured with UEFI (OVMF) instead of
BIOS. I head its about dropping OpRegion dependency in Windows guest driver to
support that configuration

Thank you.

20.04.2019, 05:14, "Yuan, Hang"
<hang.yuan@intel.com>:
> Could you elaborate a bit more about the first two features? On UEFI VMs
support (Windows guest), do you mean enabling Windows VM display on UEFI?
Thanks.
>
>>  -----Original Message-----
>>  From: iGVT-g [mailto:igvt-g-bounces@lists.01.org] On Behalf Of Alex
Ivanov
>>  Sent: Friday, April 19, 2019 2:51 AM
>>  To: libvirt-users@redhat.com; igvt-g@lists.01.org;
virt-tools-list@redhat.com
>>  Subject: [iGVT-g] GVT-g - suboptimal user experience
>>
>>  Hi.
>>  In the current state of gvt-g the user experience is suboptimal.
>>  So my question is what are the ETAs for following features:
>>  1. Accelerated virt-manager console using gvt-g device 2. Custom
resolutions
>>  or dynamic resolution 3. UEFI VMs support (Windows guest) Thanks.
>>  _______________________________________________
>>  iGVT-g mailing list
>>  iGVT-g@lists.01.org
>>  https://lists.01.org/mailman/listinfo/igvt-g

Hi,
> 1. Its about local display / dmabuf feature. Currently user needs to
> use virt-viewer tool. But virt-manager already incorporates graphical
> console. It would be nice if it could support accelerated gvt-g local
> display.
virt-manager and virt-viewer essentially run the same code spice client
code: both are using the spice-gtk widget.  So, if that isn't working
(didn't verify) it should be rather easy to wire up.
> Also virt-manager should allow to use mdev videocard alone, without
> requiring to use additional primary virtual adapter like QXL or Cirrus
That works fine, you can just set the video card model to "none" in
libvirt xml.  Not sure whenever virt-manager allows to do that via GUI,
maybe you have to use "virsh edit" instead.
> 2. Currently with gvt solution there few mdev devices each with fixed
> resolution. User should be able to configure own resolutions for
> devices at host without additional manipulations in guest. Ideally,
> resolution should be dynamic and change appropriately on resize of
> virt-manager window
Works.  Needs cutting edge software versions though (qemu 4.0 and linux
kernel 5.1 on the host).
> 3. This is about windows guest and VM configured with UEFI (OVMF)
> instead of BIOS. I head its about dropping OpRegion dependency in
> Windows guest driver to support that configuration
That one is for intel to answer ...

cheers,
  Gerd

> -----Original Message-----
> From: Gerd Hoffmann [mailto:gerd@kraxel.org]
> Sent: Thursday, May 9, 2019 3:28 PM
> To: Alex Ivanov <gnidorah@ya.ru>
> Cc: Yuan, Hang <hang.yuan@intel.com>; libvirt-users@redhat.com; igvt-
> g@lists.01.org; virt-tools-list@redhat.com
> Subject: Re: [iGVT-g] GVT-g - suboptimal user experience
> 
>   Hi,
> 
> > 1. Its about local display / dmabuf feature. Currently user needs to
> > use virt-viewer tool. But virt-manager already incorporates graphical
> > console. It would be nice if it could support accelerated gvt-g local
> > display.
> 
> virt-manager and virt-viewer essentially run the same code spice client
> code: both are using the spice-gtk widget.  So, if that isn't working
(didn't
> verify) it should be rather easy to wire up.
> 
> > Also virt-manager should allow to use mdev videocard alone, without
> > requiring to use additional primary virtual adapter like QXL or Cirrus
> 
> That works fine, you can just set the video card model to "none"
in libvirt
> xml.  Not sure whenever virt-manager allows to do that via GUI, maybe you
> have to use "virsh edit" instead.
> 
> > 2. Currently with gvt solution there few mdev devices each with fixed
> > resolution. User should be able to configure own resolutions for
> > devices at host without additional manipulations in guest. Ideally,
> > resolution should be dynamic and change appropriately on resize of
> > virt-manager window
> 
> Works.  Needs cutting edge software versions though (qemu 4.0 and linux
> kernel 5.1 on the host).
> 
> > 3. This is about windows guest and VM configured with UEFI (OVMF)
> > instead of BIOS. I head its about dropping OpRegion dependency in
> > Windows guest driver to support that configuration
> 
> That one is for intel to answer ...
[Yuan, Hang] Thanks for your answers, Gerd! On dropping OpRegion dependency, we
raised the request to Windows gfx driver team but so far not in their plan yet.

So it looks like this solution is currently not very user friendly. Perhaps
Intel could get Red Hat help to make first-class support of gvt feature in
virt-manager?

09.05.2019, 10:28, "Gerd Hoffmann"
<gerd@kraxel.org>:
>   Hi,
>
>>  1. Its about local display / dmabuf feature. Currently user needs to
>>  use virt-viewer tool. But virt-manager already incorporates graphical
>>  console. It would be nice if it could support accelerated gvt-g local
>>  display.
>
> virt-manager and virt-viewer essentially run the same code spice client
> code: both are using the spice-gtk widget. So, if that isn't working
> (didn't verify) it should be rather easy to wire up.
Currently it doesn't work. Only following methods do work:
https://blog.bepbep.co/posts/gvt/
>
>>  Also virt-manager should allow to use mdev videocard alone, without
>>  requiring to use additional primary virtual adapter like QXL or Cirrus
>
> That works fine, you can just set the video card model to "none"
in
> libvirt xml. Not sure whenever virt-manager allows to do that via GUI,
> maybe you have to use "virsh edit" instead.
It doesn't allow it via GUI.
>
>>  2. Currently with gvt solution there few mdev devices each with fixed
>>  resolution. User should be able to configure own resolutions for
>>  devices at host without additional manipulations in guest. Ideally,
>>  resolution should be dynamic and change appropriately on resize of
>>  virt-manager window
>
> Works. Needs cutting edge software versions though (qemu 4.0 and linux
> kernel 5.1 on the host).
Ok, I will test it with later versions someday. However I assume dynamic
resolution (on window resize) is not supported?
>
>>  3. This is about windows guest and VM configured with UEFI (OVMF)
>>  instead of BIOS. I head its about dropping OpRegion dependency in
>>  Windows guest driver to support that configuration
>
> That one is for intel to answer ...
>
> cheers,
>   Gerd