[vfio-users] [FEEDBACK NEEDED] Additionnal tips to include in the Arch wiki article

[Okky Hendriansyah]

On Mon, May 9, 2016 at 8:56 AM, Nicolas Roy-Renaud <
nicolas.roy-renaud.1 at ens.etsmtl.ca> wrote:

> Hello everyone.
>
> I've spent the last month or so reworking the pci passthrough article on
> Arch Wiki <https://wiki.archlinux.org/index.php/PCI_passthrough_via_OVMF>,
> and I'm getting to the point where I'm starting to be pretty satisfied with
> it personally. While there are still a few things I need to adress,
> however, I'd like to get some feedback on the current article, as well as
> possibly some topics that could be worth elaborating. Since I'm sure most
> of those who are reading the mailing list have had trouble setting up their
> passthrough at one point or another, I'm sure you all have some issue
> you've managed to solve that you would like to share with others, so I'll
> be collecting those and try to see if it makes sense to include them in
> there.
>
> Here's a few of the common gotchas I'm already planning to add in the near
> future :
>
>    - Using identical GPUs on the guest and host machines (Garland Key had
>    that issue a while ago, and this requires some binding scripts)
>    - Passing through a device that cannot reset properly (I've had that
>    problem with my soundcard and still need to figure out a solution)
>    - Actually go into details about what the ACS override patch does and
>    how to configure it
>    - MSI interrupts for the GPU soundcards
>
> Also, I have a few questions for AW specifically, since he mentions a few
> interesting in his blog that could't find all that much info on and could
> use some clarification, namely :
>
>    - What impact do hugepages have on host and guest performance?
>    - How exacly does host-passthrough affect performance?
>    - Is there an easy way to test if a device can reset properly without
>    a host reboot or not? I know a few pci devices have a /reset node to
>    indicate this, but recent issues with my integrated soundcard and the
>    reocurring complaints about AMD cards are making me consider adding a
>    section about that in the article.
>
> - Nicolas
>
Hi Nicolas,

Appreciate your work on this. I would like to propose other things about
VFIO/OVMF GPU passthrough that in my opinion should be included in the wiki
as well:

1.) Avoid transparent hugepages and enable explicit hugepages (hugepages
size variance 2 MB, 1 GB)

2.) Kernel configuration to improve performance like 1000 Hz clock
resolution, Voluntary Preemptible Kernel, halt_poll_ns

3.) How to passthrough host primary GPU (if possible, :) )

4.) CPU family that should be use if you want to avoid ACS patches [1]

5.) APIC-V vs Hyper-V VAPIC on recent Intel CPU

6.) Issues on upgrading Windows 10, which is similar to your core2duo
workaround [2][3]

7.) In case someone interested in passing through GPU to a Mac OS X guest
[4]


[1] http://vfio.blogspot.com/2015/10/intel-processors-with-acs-support.html
[2] https://www.redhat.com/archives/vfio-users/2015-November/msg00164.html
[3] http://ubuntuforums.org/showthread.php?t=2289210
[4]
http://www.insanelymac.com/forum/topic/309087-insanely-fast-virtual-mac-qemu-ovmf-clover-and-native-graphics/

Best regards,
Okky Hendriansyah
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/vfio-users/attachments/20160510/23073c74/attachment.htm>