RE: iommu=no-igfx

["Tian, Kevin" <kevin.tian@xxxxxxxxx>]

> From: Jan Beulich <jbeulich@xxxxxxxx>
> Sent: Wednesday, May 20, 2020 7:11 PM
> 
> On 11.05.2020 19:43, buy computer wrote:
> > I've been working on a Windows 10 HVM on a Debian 10 dom0. When I
> was first
> > trying to make the VM, I was getting IOMMU errors. I had a hard time
> > figuring out what to do about this, and finally discovered that putting
> > iommu=no-igfx in the grub stopped the errors.
> >
> > Unfortunately, without the graphics support the VM is understandably
> slow,
> > and can crash. I was also only now pointed to the page
> > <https://xenbits.xen.org/docs/unstable/misc/xen-command-
> line.html#iommu>
> > which says to report any errors that get fixed by using iommu=no-igfx.

what is the platform and linux kernel version in this context?

> 
> Thanks for the report. For context I'll quote the commit message of
> the commit introducing the option as well as the request to report
> issues fixed with it:
> 
> "As we still cannot find a proper fix for this problem, this patch adds
>  iommu=igfx option to control whether Intel graphics IOMMU is enabled.
>  Running Xen with iommu=no-igfx is similar to running Linux with
>  intel_iommu=igfx_off, which disables IOMMU for Intel GPU. This can be
>  used by users to manually workaround the problem before a fix is
>  available for i915 driver."
> 
> This was in 2015, referencing Linux >= 3.19. I have no idea whether
> the underlying driver issue(s) has/have been fixed. The addresses
> referenced are variable enough and all within RAM, so I'd conclude
> this is not a "missing RMRR" issue.

Variable enough but not within RAM. From E820:

(XEN)  0000000100000000 - 0000000871800000 (usable)

But the referenced addresses are way higher:

(XEN) [VT-D]DMAR:[DMA Read] Request device [0000:00:02.0] fault 
addr 76c615d000, iommu reg = ffff82c000a0c000
(XEN) [VT-D]DMAR: reason 06 - PTE Read access is not set

> 
> Cc-ing the VT-d maintainer for possible insights or thoughts.
> 
> Jan

I don't have other thoughts except the weird addresses. It might be
good to add some trace in dom0's i915 driver to see whether those
addresses are intended or not.

Thanks
Kevin