vfio-pci.txt

How to use:
1. Check if VT-d in BIOS is turned on for Intel, AMD-Vi for amd
2. Add intel_iommu=on/amd_iommu=on to kernel line, separately
3. Load vfio-pci, vfio, vfio_iommu_type1 modules
4. Currently, for Intel VT-d1 system and AMD system without interrupt remapping support, allow_unsafe_interrupts should be enabled by:
   # echo 1 > /sys/module/vfio_iommu_type1/parameters/allow_unsafe_interrupts
5. Assume user wants to access PCI device 0000:06:00.0. Check which IOMMU group this PF is in by:
   # readlink /sys/bus/pci/devices/0000:06:00.0/iommu_group
6. Unbind this PF from host driver and bind it to vfio-pci:
   # echo "8086 10fb" > /sys/bus/pci/drivers/vfio-pci/new_id 
   # echo 0000:06:00.0 > /sys/bus/pci/devices/0000\:06\:00.0/driver/unbind
   # echo 0000:06:00.0 > /sys/bus/pci/drivers/vfio-pci/bind
7. Check what other devices are in the same group as PF mentioned above, unbind all of them and bind to vfio-pci:
   # ls -l /sys/bus/pci/devices/0000:06:00.0/iommu_group/devices
8. Launch a guest by assigning PF:
   /usr/libexec/qemu-kvm ... -device vfio-pci,host=06:00.0,id=pf
9. Bind back to its initial driver:
   # echo "8086 10fb" > /sys/bus/pci/drivers/ixgbe/new_id 
   # echo 0000:06:00.0 > /sys/bus/pci/drivers/vfio-pci/unbind
   # echo 0000:06:00.0 > /sys/bus/pci/drivers/ixgbe/bind