[vfio-users] Passing intel_iommu=on in grub causes boot to immediately lock on two different X58 LGA1366 MB.

[Alex Williamson]

On Mon, May 23, 2016 at 3:20 AM, Brian Yglesias <
brian at atlanticdigitalsolutions.com> wrote:

> Sorry for the schizophrenic posts.
>
> Updating the BIOS reset it to default, which returned VT-d to the default
> setting of 'disabled', and that is the only reason I was able to boot.
>
> I tried the new card in both motherboards, and the result is the same.
>
> So, just to update my previous response:  VT-d remains broken on both the
> Rampage II Gene and GA-EX58-UD5 (probably only for Linux, not sure).
>
> Idk if it would be better if a moderator just deleted this and the last
> post, just to avoid any confusion.
>
> At this point I am back at square one.  I don't want to rely on the list I
> found over at the Arch Linux forum, as that got me the Gigabyte board.  The
> info on the Xen wiki just says VT-d is supported on X58, but my experience
> with two higher end boards strong suggests that's an overstatement, if not
> just mostly wrong.
>
> If anyone knows of any half decent LGA 1366 board that supports VT-d
> correctly in Linux, then I would really appreciate the pointer.  It would
> save me a a lot of time and money, both of which I've been been spending
> thus far with no result.
>
> I'll settle for /anything/ with two x16 slots that works.
>

I have a Lenovo Thinkstation S20 (4157CTO) that works reasonably well for
X58-based VT-d work.  Complete systems on ebay are cheap and the power
supply and maybe even motherboard appear to be standard form factors (have
never tried swapping them).  A couple caveats on X58 systems, 1) interrupt
remapping is generally broken on these systems due to processor hardware
errata, it will be disabled and you'll need to
use allow_unsafe_interrupts=1 on the vfio_iommu_type1 module (not a big
deal if you minimally trust your guests), 2) ICH10 does not have ACS
support and will never have quirks, due to being too old for Intel to care
to investigate enabling isolation.  Processor root ports, such as the x16
slots, do have ACS.  Caveat specific to the S20: the onboard NIC is
attached to the ICH10 where surprise hotplug is enabled on the root port
(yes, even though it's soldered onto the motherboard), the NIC (tg3) has no
reset mechanism, so we do a PCI bus reset, which can trigger said surprise
hotplug.  This seems like something I'm constantly fixing upstream, but
it's attached to the ICH10 so these are best left for the host anyway.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/vfio-users/attachments/20160523/1a8866e0/attachment.htm>