[vfio-users] nVidia Code 43

[Alan Jenkins]

Hmm I am starting to think that this is not nVidia's monkeying with the
driver.  I just tried installing driver 337.50 which I believe is from
before they "accidentally" added the bugs that cause us problems and I am
still getting Code 43.

With regards to the ignore_msrs option for KVM it has been enabled during
all of my recent testing as I read that that helps.

To summarise what I have tried so far:

* Double checked that PCI-E power connectors are plugged into the card
properly (they were and the card was working fine in windows but checked
anyway to rule it out).

* Tried both q35 and i440FX chipsets.

* Installed nVidia's 337.50 beta driver which I believe was from before the
restrictions.

Is there any possibility that there is some quirk with passing through a
gtx690 as it is a dual GPU card? Has anyone else had success with one?

Thanks,

Alan Jenkins

On Fri, 11 Dec 2015 10:04 Hristo Iliev <hristo at hiliev.eu> wrote:

> On Fri, 11 Dec 2015 08:08:41 +0000
> Alan Jenkins <alan.james.jenkins at gmail.com> wrote:
>
> > Hey guys,
> >
> > I have spent the last 2 days experimenting trying to get my GTX690 passed
> > through to a Windows 10 guest and since succeeding have been getting
> > nothing but Code 43 errors.
> >
> > In Windows 10 I am currently using the latest nVidia driver (359.06).
> >
> > My libvirt xml is here http://pastebin.com/KY9QWhcu note that there is a
> > device commented out from passthrough, that is a PCI express SSD which I
> > will be trying to make work but for the time being I am focusing on
> > getting getting the GTX 690 to function (remember it is a dual GPU card
> > which is why there is so much forwarded).
> >
> > I am running Arch Linux with the linux-vfio-lts kernel from AUR as I
> > discovered the hard way that there is currently a bug in the normal
> > kernels that causes passthrough to hang the VM while it is loading
> > Windows and cause the host machine to stutter. The exact kernel version
> > is: 4.1.13-1-vfio-lts #1 SMP PREEMPT Thu Dec 10 13:21:54 GMT 2015 x86_64
> > GNU/Linux.
> >
> > My libvirt version is: 1.3.0-1.
> > My Tianocore uefi is from this RPM:
> > edk2.git-ovmf-x64-0-20151209.b1361.gfaec499.noarch.rpm
> >
> > Have nVidia stepped up their efforts to block us from using vga
> > passthrough or is there something I have missed in my XML that could be
> > indicating to the driver that it is running in a VM which then causes it
> > to Code 43 me?
> >
>
> Hi Alan,
>
> Besides switching to a 440FX machine type, try also commenting out the
> following:
>
>   <hyperv>
>     <relaxed state='off'/>
>     <vapic state='off'/>
>     <spinlocks state='off'/>
>   </hyperv>
>
> My Windows 10 VM (working XML config attached) works fine without it and I
> regularly update the driver via NVIDIA Experience. For the latter to work,
> add to /etc/modprobe.d/kvm.conf the following (create the file if it
> doesn't exist):
>
> options kvm ignore_msrs=1
>
> While not strictly needed for the NVIDIA driver to work, it makes some
> programs that probe for various CPU features, including NVIDIA
> Experience, happy.
>
> Regards,
>
> Hristo
>
> > Thanks for any help in advance,
> >
> > Alan Jenkins
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/vfio-users/attachments/20151211/d7080877/attachment.htm>