[vfio-users] choosing motherboard/chipset, hardware compatibility lists

[Alex Williamson]

On Thu, Mar 3, 2016 at 6:36 AM, Daniel Pocock <daniel at pocock.pro> wrote:

>
>
> On 03/03/16 13:15, Alex Williamson wrote:
> > On Thu, Mar 3, 2016 at 5:33 AM, Daniel Pocock <daniel at pocock.pro
> > <mailto:daniel at pocock.pro>> wrote:
> >
> >
> >
> >     After the problems with the Intel 55x0 chipset, I've been looking at
> how
> >     to choose something that is more likely to work
> >
> >     I came across some lists maintained in various places:
> >
> >     https://en.wikipedia.org/wiki/List_of_IOMMU-supporting_hardware
> >       (mentions the C612 chipset but not the earlier C602?)
> >
> >
> https://docs.google.com/spreadsheets/d/1LnGpTrXalwGVNy0PWJDURhyxa3sgqkGXmvNCIvIMenk/edit?pref=2&pli=1#gid=0
> >
> >     http://support.citrix.com/article/CTX131385
> >     (this says HP Z800 is supported, but maybe they were testing one with
> >     the newer 5520 C2 chipset)
> >
> >
> https://www.vmware.com/resources/compatibility/search.php?deviceCategory=io
> >
> >     http://www.odin.com/support/extreme/
> >        (mentions a few chipsets)
> >
> >     Are there any others that people consider useful for KVM VGA
> >     passthrough?
> >
> >
> >  http://vfio.blogspot.com/2015/10/intel-processors-with-acs-support.html
> >
> > VT-d is more in the processor than the chipset these days.  Picking the
> > right processor will automatically pick the right chipset, which will be
> > X79/X99 (or equivalent C-series) based.
> >
>
>
> Thanks for this feedback
>
> X79 is Patsburg, appears to be equivalent[1] to C602  (HP Z420, Z620, Z820)
>
> X99 is Wellsburg, appears to be equivalent[2] to C612 (HP Z440, Z640, Z840)
>
> Your blog comments on the features moving from chipset to CPU.  Does
> that mean that BIOS and motherboard manufacturer have less impact on
> success as well, or these are still strong factors?
>
> Is there anything you would consider to be a compelling reason to use
> Wellsburg-based chipsets, or if I locate a box from the Patsburg era
> that should suffice?
>

I suspect that since we've gotten past vendors actually providing an option
to enable VT-d, there's very little they can do to screw it up, other than
place RMRR requirements on devices (which is generally not a problem if you
don't have an HP iLO).  I have a Z820 that works just fine for device
assignment and I test it regularly with Quadro VMs.  I also have an X58
(Tylersburg ~5520) which is subject to the interrupt remapping issue and it
also works just fine, so long as you opt-in to the isolation risks
associated with lack of interrupt remapping.  I don't know of any feature
that you'll see specifically related to VT-d in Patsburg vs Wellsburg, both
have quirks for exposing ACS-equivalent isolation in the PCH root ports,
though Wellsburg came later (v4.0) and therefore might need a newer kernel
than Patsburg (v3.16).  Thanks,

Alex
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/vfio-users/attachments/20160303/b610dbd8/attachment.htm>