Re: BUG: libxenlight fails to grant permission to access Intel IGD Opregion

[Chuck Zmudzinski <brchuckz@xxxxxxxxxxxx>]

On 3/11/22 3:09 AM, Jan Beulich wrote:
> On 11.03.2022 06:01, Chuck Zmudzinski wrote:
>
> > Here is the patch that fixes the bug on Debian sid/Xen 4.16:
> As to an actual patch for us to take - please see
> docs/process/sending-patches.pandoc for the formal requirements.
> (Note this was recently introduced, so you won't find it in the
> 4.16 sources. But your patch wants to be against latest staging
> anyway.)
>
> Jan
OK, I took a look at the process and I also studied this
issue more closely, and my conclusion is that I would not
recommend fixing this old bug now until we have a better
idea about how good our current tests for the Intel IGD are.

AFAICT, if our tests for the Intel IGD result in a false positive,
then hvmloader will map three pages in the guest for the
IGD opregion, but the mapped memory would certainly
not be the expected IGD opregion if the device is not actually
an IGD or GPU with an opregion. In such a case, we would be
mapping three pages of unexpected memory to the guest. So before
proposing a patch that would fix this bug but have the unintended
consequence of allowing access to unexpected memory in the case
of a false positive detection of an Intel IGD, I will first spend some
time deciding if a more accurate and reliable test is needed to
determine if a PCI device with class VGA and vendor Intel actually
has an IGD opregion. Once I am confident that the risk of a false
positive when testing for the Intel IGD is acceptably low , then I
would consider submitting a patch that fixes this bug.

Our tests check if the PCI device has class VGA and that the
vendor is Intel, and we also check if the gfx_passthru option
is enabled. Those tests are applied both in hvmloader
and in libxenlight to decide about mapping the IGD opregion
to the guest and informing Qemu about the mapped address.
I don't think these tests for the Intel IGD account for recent
developments such as newer discrete Intel GPUs that
might not have an IGD opregion, nor do they account for
older Intel IGDs/GPUs that also might not have an IGD opregion.

I think some time is needed to look at the i915 Linux kernel driver
code to more precisely identify the devices that need access
to the IGD opregion. Other devices either are not compatible
with the feature of VGA passthrough or do not need to have
access to the IGD opregion.

With this information, a patch can be developed that will more
accurately determine when the guest needs access to the IGD
opregion. With such a patch committed in Xen, I would be more
comfortable submitting a fix for this bug.

Regards,

Chuck