Xen project Mailing List

We solved this problem recently. The issue we found was that the USB interfaces were being picked up and bound to the host OS before the pciback module had loaded. This meant that the oddball module was loading, trying to bind its configured devices and finding they were already taken.

Which version of Xen are you running? We found we couldn't make pass through work until we upgraded to 4.2.0. Also,.it may sound silly but check in the bios that all the virtualization features are enabled. Some of the boards we have come out the box with them on, some with them off. It seems to depend on whether the board was intended for a server or a workstation.

Try this test (assuming you are running the xl toolkit - if not, modify the command to be appropriate for xm):

Shut down the windows guest with xl shutdown guest_name
Manually bind the PCI devices: xl pci-assignable-add 00:14.0
xl pci-assignable-add 00:1a.0
xl pci-assignable-add 00:1d.0
Restart the guest with xl create filename

If the USB interfaces magically appear in Windows, this confirms that the y are getting bound to the host early.

One is to rebuild the host kernel and compile the Xen stuff into it so pciback is in the kernel and gets there first. This, I believe, is the "proper" way to do it.

The other I know of is to modify the xendomains script to pci-assignable-add the BDFs as above before starting the guests. This isn't so nice but you don't have to build a new kernel.

Sorry for the bad layout of this email. Written on my phone with fat fingers on a moving train :-)

Philip Wernersbach <pwernersbach@xxxxxxxxxxxxxxxxxxxx> wrote:

Paul,

Thanks for taking the time to help. The host operating system is NetBSD 6.0. ÂThere is one Windows XP guest, and the output of lspci -vvnn is atÂhttp://pastebin.com/zPNdnnjK. We are trying to pass through the USB controllers on the machine, which should be PCI addresses 00:14.0, 00:1a.0, and 00:1d.0.

Thanks,

Philip Wernersbach
Software Implementation Division
Jacobs Automation
On Wed, May 15, 2013 at 3:49 AM, Paul Stimpson <paul.stimpson@xxxxxxxxxxxxxxxxx> wrote:
Hi Philip,

I run a consultancy and I'm just working with a name-you-know company to deliver a xenized appliance that uses IOMMU.

We found that getting IOMMU working was one if the more challenging parts of the project.

Please will you tell me what the host operating system is (including the version), how many guests there are, what the guests' operating systems are, what the piece of hardware you are trying to pass is and the output of the command lspci?

Best regards,
Paul.
Philip Wernersbach <pwernersbach@xxxxxxxxxxxxxxxxxxxx> wrote:
Hello Everyone,

My company is in the process of implementing Xen virtualization. We have managed to get Xen up and running, and have worked through all of the kinks in our set up, except for PCI passthrough. PCI passthrough doesn't work, and I suspect it's because Xen is not using the IOMMU in the computer, even though we built the computer with components that support VT-d.

Here are the computer components we used:
Motherboard:ÂIntel BOXDH77DF
CPU:ÂIntel Core i5-2310 (Sandy Bridge - BX80623I52310)

I also booted Xen up with the "iommu=verbose" parameter andÂran "xl dmesg" on the computer, and uploaded the output here:Âhttp://pastebin.com/rF0kyftt. I tried running Xen with the "iommu=required" flag, and it failed because it said it couldn't use the IOMMU. If someone work with us to figure out why Xen doesn't use the IOMMU, it would be very appreciated.

Thanks.

Philip Wernersbach
Software Implementation Division
Jacobs Automation
Xen-users mailing list
Xen-users@xxxxxxxxxxxxx
http://lists.xen.org/xen-users
--
Sent from Kaiten Mail on Android. Please excuse my brevity.

Re: [Xen-users] Xen Reports no IOMMU, but Hardware Supports It