On Sun, Mar 1, 2026 at 11:29 AM Jason Gunthorpe <[email protected]> wrote: > > On Fri, Feb 27, 2026 at 11:25:01AM -0700, Alex Williamson wrote: > > > There's obviously a knee jerk reaction that moving PF drivers into > > userspace is a means to circumvent the GPL that was evident at LPC, > > even if the real reason is "in-kernel is hard". > > Given we already have GPL licensed kernel drivers for the PFs it > doesn't seem like a reasonable worry to me to fret about some cut down > version of a kernel driver running in userspace. > > Further, let's be honest here, the people most interested in all of > this are doing it to support their proprietary VMMs. I'm pushing that > we must have at least a reference implementation in qemu before the > kernel parts should be merged..
Yes, this is why we built VFIO selftests with the driver framework [1]. We wanted to ensure Live Update support could be well tested upstream independent of VMM adoption and without requiring complex orchestration. Google is also working on upstreaming Live Update support into Cloud Hypervisor and the kernel pieces land. [1] https://lore.kernel.org/kvm/[email protected]/
