Getting modern virtio-pci support on kvmtool to not break virtio-blk

Getting modern virtio-pci support on kvmtool to not break virtio-blk

[Zach Reizner]

Good Afternoon,

I've been trying to make Sasha Levin's "kvmtool: add support for
modern virtio-pci" patch work on master and it doesn't seem to behave.
The patch itself applies cleanly and the kernel does boot, but as soon
as the kernel tries to read from the virtio-blk device, the kvmtool
side virtio-blk device never responds to the request. From generously
sprinkled prints, I know that that the virtio-blk device is being
notified of work on the vring, but when it actually checks for any
(using virt_queue__available) it decides there is none available. In
trying to discover why the virtio queue is empty, I've come up on a
wall.

Seeing as I'm working with an apparently unmerged patch, perhaps this
is a known issue. Maybe even somebody knows what I'm doing wrong and
how to fix it. I've also been curious why this patch is still unmerged
after a year (to this exact day day).

A little background: I'm trying to get modern virtio-pci working
because I want to add basic virtio-gpu support to kvmtool, and the
linux guest kernel module for that requires modern virtio-pci.
For reference, I'm starting kvmtool with

$ lkvm run --console virtio -m 4096 -c 1 --disk
/data/scratch/archkvm.squashfs --kernel arch/x86/boot/bzImage"

where archkvm.squashfs is a custom userspace bootstrapped from arch
linux (although I doubt that matters as the kernel doesn't get to
mounting the rootfs). This command works perfectly fine with modern
virtio-pci disabled. My kernel config is pasted here:
http://hastebin.com/tufelatuji.txt .

Thanks,
Zach Reizner