خرید بک لینک

I'm trying to make use of SR-IOV on my network card, and it appears that because Intel didn't implement ACS on the Xeon E3-1245 v5 (Skylake), the network cards I'm trying to assign to a guest are in the same IOMMU group as a PCI Express Root Port:

Code:

IOMMU group 6        
        00:1d.0 PCI bridge [0604]: Intel Corporation Sunrise Point-H PCI Express Root Port #9 [8086:a118] (rev f1)
        07:00.0 Etheet controller [0200]: Intel Corporation 82576 Gigabit Network Coection [8086:10c9] (rev 01)
        07:00.1 Etheet controller [0200]: Intel Corporation 82576 Gigabit Network Coection [8086:10c9] (rev 01)
When I enable vfs in the igb module, the virtual interfaces also show in group 6. When I try to assign one them to a kvm guest, I get the error:

Code:

vfio: error, group 6 is not viable, please ensure all devices within the iommu_group are bound to their vfio bus driver.
If I assign 07:00.01 and 07:00.1 to vfio-pci, then I lose the virtual interfaces and still get the error since the Root Port at 00:1d.0 isn't bound to vfio.

I believe the solution to this is to use the acs_override keel patch. However I can't find a link to one that applies for the 4.4 keel in Ubuntu 16.04. Is it already included? If not, does anyone have a link to one that works? Or is there some other quick or workaround I can use to properly assign this to a VM? I won't be using these cards on any other VMs, so I don't care if there's not proper DMA isolation between guests since there's only 1 guest.

Right now I have regular PCI passthrough working fine, but the DMA copy operations are killing my CPU (maxes out 2-3 cores transferring data at 150 mbps, with the interrupts associated with vfio-msix being the culprit). SR-IOV will eliminate this which is why I want to use it.

برچسب: نویسنده: استخدام کار تاريخ: سه شنبه 8 تير 1395 ساعت: 2:35

صفحه بندی