[vfio-users] How would you estimate my probability of a successful passthrough on this hardware?

[Alex Williamson]

On Wed, 31 May 2017 10:24:17 -0700
Nick S <nick.kvmhv at gmail.com> wrote:

> Hi vfio-users,
> 
> I am looking for expert opinions whether I will be able to successfully
> passthrough a GPU on the following hardware:
> 
> 1) CPU: i7-3770 (not K)
> 2) Motherboard: ga-b75m-d3h Gigabyte that claims to support VT-x and VT-d
> but I was not able to find any success stories
> 3) Using Intel Graphics for host and either AMD or Nvidia GPU for guest. I
> have plenty of options here, aiming for OVMF setup may be with Nvidia
> 1060gtx.
> 
> I already have a setup based on a Haswell-E CPU and trying to set up a
> second PC on some older hardware I have. As I understand, this setup will
> require ACS override that I did not have to apply for my Haswell-E
> configuration. So, a couple of questions if it looks feasible:
> 
> 1) What is the best place to find the ACS override patch for 4.10 kernel?
> 2) Should I expect any other complications compared to a relatively
> straight forward Haswell-E setup where I also have OVMF guests and BIOS
> host?

I don't see any reason you'd need to use the ACS override patch here.
The block diagram on page 6 of the manual shows that the x16(x16) slot
is the only slot connected directly to the CPU.  All of the onboard I/O
is shown from the PCH.  Therefore lack of ACS on that one slot only
means that any card you plug into the x16 top slot can't be split among
multiple users... but you don't intend to do that anyway. If you intend
to use the conventional PCI slots, they can't be split regardless
because the IOMMU cannot distinguish transactions from either of those
slots or anything plugged into them.  The bottom x16(x4) slot is
sourced from the PCH so will be handled by existing quirks.  This is
almost identical to the layout of the board I use in my example setup
on the blog.  Thanks,

Alex