[vfio-users] Checking for EFI and general questions

[Jens Zimmermann]

Thank you again for your responses! I took some time to analyse your
answers and do further research which is why my reply is a bit delayed.

So first regarding my GPU:

QEMU can use a file to expose as a ROM for the card, see the romfile=
> option of vfio-pci.
>

I am not sure I understand this correctly. Does this mean QEMU assigns a
UEFI ROM of another 750 Ti on my card when it launches the VM? So I don't
have to flash my GPU even though it does not support UEFI? If so, can it
cause any problems because it's not the original ROM or because of the
implementation in QEMU?
I did spent quite a lot of time searching for a 750 Ti UEFI ROM, but I
could not find any (but maybe I am just too stupid to use Google). I am
wondering about this since some people in the Google docs list claim that
they have managed to get it working using OMFV/UEFI with a 750 Ti.

You can also add an emulated GPU and install the guest drivers which
> eliminate the need of GOP, you'll just have to boot the guest without video
> output until the driver loads.
>
Is that the same as the romfile method?
I am starting to become a bit confused now... Is there a simple and good
way of using GPU passthrough with my 750 Ti or do I need to buy a new card
that has native UEFI support?


As for general hardware:


>
> Nvidia cards often lack the EFI driver in their ROM, AMD cards often
> have it broken or don't have it at all..
> Motherboards sometimes have bugs in ACPI(see not so long ago case with

intel desktop board and pci config space or an old case with broken
> ACPI table on AMD FX platform), sometimes there are bugs in
> IOMMU(FM2[+]), sometimes they are just made the way you can't really
> assign anything(modern intel platforms with tons of system devices in
> the same group as the GPU).
>
> From what I observe, it's just a gamble.
>

I realise that you can't be 100% sure beforehand, what I am looking for is
a setup that will most likely work.
As I understand it using UEFI/OMFV is the preferred method compared to
non-UEFI/SeaBIOS, right?

Regarding the CPU, it needs to have vt-d of course.
I read the article about IOMMU and ACS. So there are devices which are
grouped in IOMMU groups, I can only passthrough a whole group. This can
cause problems if there is a group where I want to passthrough one but not
all devices. A way to fix this is a patched kernel like this
https://aur.archlinux.org/packages/linux-vfio/. This is no problem for
Haswell-E. Correct?

So if I want to have one Windows-VM (GPU, audio, USB passthrough), is it
like:
a) there will be no problems with IOMMU groups on any i5, i7 or Xeon E3?
b) there can be problems with IOMMU groups using any i5, i7 or Xeon E3 so
patching would be required; luck related (depending on the hardware)?
c) there definately will be problems with IOMMU groups on any i5, i7 or
Xeon E3? I definately need to apply an ACS patch or use Haswell-E/Xeon E5
instead?


Other than that, thanks for the explanations and links (yes, I am German
:-) ), I think I understand a bit more now. What I don't understand, why
can I use the nvidia 210 only with Haswell-E?

Thank you for reading! I really appreciate your help! :-)
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/vfio-users/attachments/20151124/9ef5501b/attachment.htm>