[vfio-users] IGD assignment users, please test

[Manuel Ullmann]

Hi Alex,

for me, the patch also didn’t have any impact (negative or
positive). I’ve used this on a Skylake Laptop (Lenovo T460). Actually
the passthrough tended to work out of the box with the configuration
copied from my Haswell Desktop. I do/did run into a few issues though:

1.  The patch could not be applied to qemu-2.8.0 or git, which was
unfortunate. I’m not sure against which branch or commit it was written,
but especially the hwaddr line cutting off the error message seems odd.
Manually applied the advertised changes into a new one working against
v2.8.0:
http://pastebin.com/1LJXZskq

2.  Occasionally (maybe 50% of boots) the VM will crash at Windows
loading screen. Could be related to the following thread, although I
don’t get a kernel oops.
https://www.redhat.com/archives/vfio-users/2016-December/msg00064.html
qemu crash output:
http://pastebin.com/36rnB5Zq

3.  For some reason I couldn’t allocate 6 1GB hugepages, although both
computers (Haswell desktop/Skylake laptop) have 8 GB memory and
regardless of specifying them as kernel boot parameter or echoing them
into /proc/sys/vm/nr_hugepages in initrd. Also the boot parameter seemed
to have no allocation effect at all, which was only achieved by manually
echoing into the sys node, resulting 5 pages being allocated. Slightly
confusing, that an rsynced system behaves completely different on other
hardware. Network relevant deviations were done by abusing portage’s
config protect feature and rsync’s filter lists. Anyway, works (almost)
fine with 5 hugepages.

4.  I still have trouble assigning an ALSA device to the emulated ich9
card. It works, but the Audio is distorted. I worked around this with my
Haswell by passing through the Intel PCH and using the USB DAC as common
sound device. This somehow fixed the distortion, although I still have
some popping sounds occasionally (as if the DAC was in passthrough).

On my Skylake laptop the PCH audio device is unfortunately not
isolated. Almost half of the processor is on the same bridge including
MMU and SMbus and ISA bridge. So I can’t work around in the same
manner. Passing through the DAC works with the same quality.

Could this somehow be related to the IGD being passed through and still
sharing any resources with the Audio chipset? I didn’t get resolving
tips on libvirt-users mailing list.

Best regards,
Manuel

> 	Hi Alex,
>
> 	For me it did no difference, but all in all, my case was far away from 
> getting working because of the intrusive simplefb jumping into the field.
>
> 	Thanks!
>
> 	José.
>
> On Monday 12 December 2016 15:24:40 Alex Williamson wrote:
>> Hi,
>> 
>> I think maybe we've found the issue that many of you with Skylake or
>> newer systems were seeing with IGD assignment.  There's a wraparound
>> issue with the GTT programming, as discovered by Alexander Indenbaum.
>> I still don't have access to anything newer than Broadwell, so I'm
>> interested in test reports from anyone that's currently using IGD
>> assignment or tried previously and maybe didn't get any output or
>> garbled output, whether on Skylake, Kabylake, or regression testing on
>> older IGD as well.  Please add the following patch to your QEMU build
>> and report if it breaks anything, fixes anything, or even if it stays
>> the same:
>> 
>> https://lists.nongnu.org/archive/html/qemu-devel/2016-12/msg01600.html
>> 
>> Thanks,
>> Alex
>> 
>> _______________________________________________
>> vfio-users mailing list
>> vfio-users at redhat.com
>> https://www.redhat.com/mailman/listinfo/vfio-users
>
>
> _______________________________________________
> vfio-users mailing list
> vfio-users at redhat.com
> https://www.redhat.com/mailman/listinfo/vfio-users