[vfio-users] Consolidated VFIO site

[Jonathan Panozzo]

One of the biggest challenges in building a database showing VFIO compatibility (especially with GPU assignment) is the variance in different configurations that users may have in their setups.  This becomes very evident when you start looking at the KVM VGA Passthrough spreadsheet that was linked in the other thread.  Each user may have a different kernel, different qemu version, and different patches applied, which may or may not have been configured correctly.  In addition, settings outside the OS can also have a major influence on your success with GPU passthrough (e.g. BIOS settings).  That said, we actually do have a project in the works that will permit thousands of our users to anonymously upload their hardware configurations (with their permission).  This database will greatly aide new users in selecting the right hardware for their system builds.  In particular, we should be able to glean a great deal of information regarding interoperability between various h/w components and with VFIO.  One goal is to be able to identify combinations of h/w that work correctly for various VM applications.  In addition, we’d also like to start gathering details relating to how different motherboards create IOMMU groups naturally (without the use of PCIe ACS Override).  The goal here is to provide as much information openly and transparently as we possibly can, so that users can not only validate compatibly with their existing hardware, but purchase new hardware with confidence in being able to utilize VFIO, VGA assignment, and other device types (including TV tuners, capture cards, and other devices).

How do we intend to do all this?  We’ve already laid the framework by tuning our OS for maximum compatibility with VFIO and graphics device assignment and use a very modern Linux kernel (4.1.5 at the time of this writing).  We’ve also pre-built a virtual machine (OpenELEC) which, in a single vdisk image, has support for both OVMF and SeaBIOS boot modes.  We’ve made this VM downloadable directly within our web management interface, so we are trying to remove as much complexity as we can to reach the widest user base possible, not just Linux-savvy users that are building their own kernels, etc.  Our users don’t have to worry about installing QEMU, libvirt, or learning how to use virt-manager to create and manage their VMs.  Our goal was to tailor the implementation around consumer/prosumer users that wish to build their own systems capable of taking advantage of this technology.  And we already have the web infrastructure in place to both host the backend database as well as the front-end components for this.

We also have a number of detailed guides in our wiki that point to how users can maximize their success with GPU pass through, as well as guides that show how to optimize performance with their virtual machines.  This is in addition to the video content we’re creating and putting on youtube and the assistance we are providing to users in one of our subforums, which is dedicated to KVM usage on our OS (and one of the most popular subjects is usage of VFIO):

http://lime-technology.com/wiki/index.php/UnRAID_6/VM_Management
http://lime-technology.com/wiki/index.php/UnRAID_6/VM_Guest_Support
https://www.youtube.com/channel/UC5LYki2fFiiqYnKx1xnIJbg
http://lime-technology.com/forum/index.php?board=60.0

What I’m not sure of is how our efforts could be directly intwined with the VFIO github project itself.  I can definitely say this much:  as we exit the design and proof of concept phases for this, our plan is to continue engaging with Alex directly to ensure that we’re going about this the right way with respect to identifying user configurations and validating hardware compatibility.  

Anyhow, just thought I’d throw this out there because I couldn’t agree more that building a database like this could be extremely valuable, but in order to maximize its success, variables need to be eliminated in place of constants.  And to get the maximum amount of participation, ease of use becomes a key ingredient.

Best Regards,

Jon Panozzo
Lime Technology, Inc.


> On Sep 9, 2015, at 9:19 PM, Alex Williamson <alex.l.williamson at gmail.com> wrote:
> 
> On Wed, Sep 9, 2015 at 7:34 PM, Abdulla Bubshait <darkstego at gmail.com <mailto:darkstego at gmail.com>> wrote:
> I had mentioned earlier that it would be nice to have some sort of hardware DB that shows the VFIO compatibility and I am wondering if there is interest in taking this idea further.
> 
> I am thankful for aw's blog, but beyond that getting information about VFIO is not that straight forward. The arch forum is archived and many people seem to be having difficulty finding it. Hardware data is stored on unmaintained spreadsheet not everyone knows about, and there might be people who shy away from mailing lists.
> 
> The idea I had was one sight that would provide several things.
> 
> Forum where people could seek VFIO help.
> How Tos for getting VFIO running
> HW DB where people can post their configurations and success/problems
> Information on how to get the latest qemu/libvirt on the most popular Linux distros
> There are several other ideas, but in general it is a one stop shop for all things VFIO that will cater to a wide audience. 
> 
> I am willing to donate some of my time and resources into trying to get this up and running, but I know I cannot do this alone. So I am asking if there is interest in such a concept. I admit that I do not really know the size of the VFIO community or how much demand there is for something like this. But I am throwing this out there so that if there is interest we could hopefully get the ball rolling on this.
> 
> Things I can contribute:
> 
>  - I own the vfio project on github.com <http://github.com/> (https://github.com/vfio <https://github.com/vfio>), I understand wikis can be created on the vfio.github.io <http://vfio.github.io/> project pages, maybe more
>  - I own vfio.org <http://vfio.org/> and vfio.net <http://vfio.net/>
> 
> Things I'm not willing to do - sysadmin or host a web site
> 
> This is why I've relied on blogger and redhat to host things so far.  If we can do something useful with the github project page, propose it to me and I can add you as a member.  I'm willing to point the vfio urls somewhere if there's a legitimate proposal.  Since I announced this mailing list, just shy of a month ago, we have over 100 subscribers.  Personally I've found this mailing list to be a sufficient and preferable replacement for the archlinux forums, but I know that mailing lists are a strange and wondrous artifact of the internet for many of you (as evidenced by all the top-posting).  I've had a request to register the list with Gmame for archival purpose, I intend to do that.  If there's anything else I can help with, let me know.  Thanks,
> 
> Alex
> _______________________________________________
> vfio-users mailing list
> vfio-users at redhat.com
> https://www.redhat.com/mailman/listinfo/vfio-users

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/vfio-users/attachments/20150910/131704e5/attachment.htm>