[vfio-users] Question about IGD Passthrough

[Alex Williamson]

On Mon, Jun 20, 2016 at 10:44 AM, YounJae Rho <luxflow at live.com> wrote:

> Hi!
> I applied alex's IGD(Intel graphic display) passthrough patch recently.
> First of all, thanks alex for your hard works!
>

Welcome :)  But you should really be using qemu.git + seabios.git, and
eventually just qemu.git once pc-bios is updated.


> It seems to work perfectly in Ubuntu 16.04 but problem is windows 10.
> On windows 10, all mode except safe booting doesn't boot and keep rebooting
> Law graphic mode also keep rebooting.
> First question is
> Is there anyone that have succeed in booting windows 10?
>

Yep, me.

Second question is
> What is the difference between Gerd
> patch(
> https://www.redhat.com/archives/vfio-users/2016-January/msg00035.html)
> and alex patch?
> Currently I applied only alex's IGD patch but do I have to applied Gerd
> patch also?
>

As above, use upstream.


> Third question is
> Is there no way to test UPT mode in haswell based processor?
>

Nope, according to Intel UPT needs Broadwell or newer.  You can try, but it
depends on the guest drivers, which I have no control over.


> I tested UPT mode on both windows and linux, but both said `no vga output`
>

Minimally you'll need to use the x-igd-opregion=on to get display output in
UPT mode, but remember UPT is meant to be secondary graphics in the VM only.


> I know documentation says UPT mode only is supported in Broadwell and later
> process
> But when I boot from live cd of Igvt-g
> proje)ct(
> https://01.org/igvt-g/blogs/czhou/2016/intel-gvt-g-iso-public-releas
> <https://01.org/igvt-g/blogs/czhou/2016/intel-gvt-g-iso-public-releas
> e-q12016>

e-q12016
> <https://01.org/igvt-g/blogs/czhou/2016/intel-gvt-g-iso-public-releas
> e-q12016>, it works.
> I don't know exactly but as I know Igvt-g and UPT uses same mechanism isn't
> it?
>

No.  GVT-g is a mediated virtualization for IGD that allows multiple VMs to
make use of a single IGD through hypervisor level virtualization.  We're
working with Intel and others on this as well, but it is not the same thing
as UPT.  UPT is simply GVT-d (ie. direct assignment... I can never remember
which is which in Intel lingo), where IGD doesn't have all the system
requirements that it usually does.  IOW, it's what assignment of IGD should
have always been if the chipset designers hadn't freely spread IGD
components thoughout the chipset.


> Forgive me if I ask silly question or my bad English skill annoy you
> Thanks in advance.
>
> My test environment is
> Kernel : linux 4.6.0 (almost vanilla kernel, but some code are patched for
> Ubuntu)
> QEMU : QEMU emulator version 2.6.50 (v2.6.0-1190-g482b618-dirty)
> Seabios : up to commit `splash: Skip the RGB555 mode`
> CPU : Intel E3-1265L V3 (haswell based E3 xeon)
> QEMU parameter :
> #legacy mode
> qemu-system-x86_64 -cpu host -m 4096  -realtime mlock=on -smp
> 4,sockets=1,cores=2,threads=2 \
> -machine pc,accel=kvm,usb=off,mem-merge=off \
> -nographic -no-user-config -nodefaults \
> -device vfio-pci,host=00:02.0,id=hostdev0,bus=pci.0,addr=0x2 \
> -hda /mnt/VMs/windows.qcow2 \
> -net nic,model=virtio,addr=0x04 -net user \
> -device vfio-pci,host=00:14.0,id=hostdev1,bus=pci.0,addr=0x14 \
> -device ich9-usb-uhci3,id=uhci \
> -device usb-ehci,id=ehci \
> -device nec-usb-xhci,id=xhci \
>

Make sure you're not getting a default emulated VGA device, -vga none
should do that if -nodefaults doesn't already.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/vfio-users/attachments/20160620/cf307f2a/attachment.htm>