Topic: Userspace device IO using VFIO framework
Abstract:
VFIO is a kernel framework to interact with devices from the userspace.
Currently this forms the basis of how Qemu supports PCI passthrough
of Host devices to a Guest. This can also be leveraged to create userspace
drivers for IO devices with minimal kernel intervention.
This talk focuses on how the userspace interacts with the VFIO subsystem
and how VFIO interacts with other subsystems in the kernel. Additionally, we
will also go through the existing implementations of userspace drivers that
use VFIO to interact with an IO device.
Will also talk about the future directions on how new frameworks like
iommufd are improving upon the functionality provided by the VFIO.
Outline:
VFIO Introduction, history
How VFIO interacts with IOMMU
Userspace APIs for device access
Userspace view of the device
How Qemu exploits VFIO for guest passthrough
Limitations and Future directions
Preferred Format: 25+5 mins
Bio: Linux kernel enthusiast focussing on the VFIO and device passthrough on POWER servers for the last few years, also have been contributing to SCM, qemu, libvirt and KVM.
Regards,
Shivaprasad