On 28/11/15 20:46, Eric Shelton wrote:
> Looking through the output of 'xl dmesg' on a Skylake system
> (i5-6600K), I found a number of error messages that I do not encounter
> on a Haswell-based system.  I have tried two motherboards from
> different manufacturers, with pretty much the same results.  Below are
> some of the unexpected messages:
>
> Not enabling x2APIC (upon firmware request)
> ...
> mwait-idle: does not run on family 6 model 94
> ...
> [VT-D] iommu.c:875: iommu_fault_status: Primary Pending Fault
> [VT-D] INTR-REMAP: Request device [0000:f0:1f.0] fault index 0, iommu
> reg = ffff82c000201000
> (on motherboard 1) [VT-D] INTR-REMAP: reason 22 - Present field in the
> IRTE entry is clear
> (on motherboard 2) [VT-D] INTR-REMAP: reason 25 - Blocked a
> compatibility format interrupt request
>
> This leads to a few questions:
> 1) Is there some reason x2APIC should not be enabled on Skylake?  What
> consequence, if any, is there not having x2APIC enabled?

In this case, the firmware has set the x2apic opt-out bit in the DMAR
table, indicating that Xen should not use x2apic.

You might find an option in your BIOS to undo this; there have been
enough errata in the past in this area that I would expect it to be a
tweakable.

x2apic is the extension to xapic, which permits more than 255 cpus.  So
long as you don't have that many, there isn't a specific problem with
missing x2apic mode.

> 2) Should mwait-idle be available on Skylake?

We probably need to resync the mwait driver with Linux.  It is
whitelisted on known cpu model numbers.

> 3) What about the IOMMU errors on Skylake - are they a concern?

Yes.

In both cases, PCI device f0:1f.0 is misconfigured or misbehaving.

On motherboard 1, it is delivering an interrupt for which no remapping
entry has been set up.

On motherboard 2, it is delivering an compatibility-format interrupt, as
opposed to a remapable-format interrupt.

For motherboard 2, Xen should disallow such a configuration.  Either
interrupt remapping is enabled and all devices should be configured to
issue remmapable interrupts, or interrupt remapping is disabled and
everything should be configured to issue compatibility-format interrupts.

Either way, diagnosing the problem here starts with identifying what
f0:1f.0 is.

~Andrew
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Reply via email to