[kata-dev] VFIO groups in qemu SR-IOV passthrough

Shinde, Archana M archana.m.shinde at intel.com
Thu Mar 19 06:51:28 UTC 2020


Hi Adrian,

I am surprised you are seeing a char device on the Kata side, Kata on seeing a vfio device, does a device passthrough for the device.
On the guest side, if the driver for the device has been added to the kernel, the device would be bound to the guest driver.
We don’t do a vfio assignment on the guest side. 

Perhaps, this doc explaining how SRIOV devices are passed to Kata using device passthrough will make things clearer:
https://github.com/kata-containers/documentation/blob/8d5f2f0783fc2f848514c5d2fd1991fd7b755be7/use-cases/using-SRIOV-and-kata.md

I would suggest opening an issue on github with your exact setup and logs, so that we can track it there.

Thanks,
Archana

On 3/17/20, 11:08 AM, "Adrian Moreno" <amorenoz at redhat.com> wrote:

    Hi,
    
    I am playing with kata + qemu and SR-IOV passthrough and I have some questions
    in order to understand it better.
    
    If I add, say for example "/dev/vfio/40" to the kata container, I find the same
    char device inside the container even though vfio might not even be loaded in
    the guest kernel, so I wonder:
    
    - Why is kata creating that device? I would expect kata to expose a virtual PCI
    device (backed by the vfio device mentioned), but:
       - Wouldn't it be the guest's vfio driver (if any) the one to assign the iommu
    group to the pci device?
    
    - How is kata creating the device, i.e: what is listening on it?
    
    Thanks for the help.
    
    - Adrián
    
    
    _______________________________________________
    kata-dev mailing list
    kata-dev at lists.katacontainers.io
    http://lists.katacontainers.io/cgi-bin/mailman/listinfo/kata-dev
    



More information about the kata-dev mailing list