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

[Alex Williamson]

[re-adding vfio-users]

On Wed, May 18, 2016 at 1:48 PM, Brian Yglesias <
brian at atlanticdigitalsolutions.com> wrote:

> I've attached with quiet removed (should have done that already).
>
> Nothing jumps out at me, hopefully someone else will see something.
>
> With iommu=pt in lieu of intel_iommu=on the system boots, but no iommu
> support.  If I append iommu=pt it continues to hard lock.
>
> I've included two images.
>
> It does seem to be somehow iommu related.
>
> I found this post on centos forums:
>
> https://www.centos.org/forums/viewtopic.php?t=46809
>
> That seems to be the same problem with a non-X58 board.  Unfortunately, I
> don't have the BIOS setting he appears to have used to fix it.
>
> Thanks for all the help thus far.
>
> I'm going to post on Gigabyte forums.  I've seen on both the archlinux
> forum and the former that ppl have gotten iommu working with this board,
> which is why I bought it.  Not sure what my problem could be yet.
>

I don't think that centos issue is related, that's a DMA alias issue with
broken I/O hardware.  What you're seeing is a hang in qi_submit_sync(),
which is an impossible hang according the hardware specs, but comes up a
fair bit more often than never.  AIUI, it means that magic, undocumented
bits in hardware aren't set the way they should be and you're going to need
to hope for a BIOS update, which is all but impossible on a system as old
as X58.  Good luck :-\

Alex
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/vfio-users/attachments/20160518/6b4ff683/attachment.htm>