Uploaded July 2024 | Updated September 2026, 43 minutes ago
At the start of 2023 I traded my 20-year career as a Linux sysadmin for a new life as a full time OpenZFS developer. Going great, thanks for asking!
Because fast iterative development sucks when you need to wait for a reboot after every kernel panic, I wrote quiz, a tool to make fast edit-compile-test cycles on kernel code possible. Under the hood it uses QEMU's "microvm" profile and a custom kernel config to boot from cold into the OpenZFS test suite in a couple of seconds. Its great, and I use it hundreds of times a day.
My FreeBSD-using colleagues naturally said "cool, but what about us?!" so I started looking at what it would take bring quiz to FreeBSD. The answer to that is "its complicated", and involves either adding bhyve to QEMU as a hardware virtualisation backend, or adding support for direct kernel loading support to /usr/sbin/bhyve. "Why not both?" I said, before promptly shriveling into a corn cob.
This talk will show you quiz in action, present the Linux direct boot sequence and show how I taught bhyve about it, also show how QEMU would like to work and why it doesn't quite line up with bhyve's view of the world, and hopefully show you how low-level kernel hacking for any OS can be made as simple as hacking on any other program.
At the start of 2023 I traded my 20-year career as a Linux sysadmin for a new life as a full time OpenZFS developer. Going great, thanks for asking!
Because fast iterative development sucks when you need to wait for a reboot after every kernel panic, I wrote quiz, a tool to make fast edit-compile-test cycles on kernel code possible. Under the hood it uses QEMU's "microvm" profile and a custom kernel config to boot from cold into the OpenZFS test suite in a couple of seconds. Its great, and I use it hundreds of times a day.
My FreeBSD-using colleagues naturally said "cool, but what about us?!" so I started looking at what it would take bring quiz to FreeBSD. The answer to that is "its complicated", and involves either adding bhyve to QEMU as a hardware virtualisation backend, or adding support for direct kernel loading support to /usr/sbin/bhyve. "Why not both?" I said, before promptly shriveling into a corn cob.
This talk will show you quiz in action, present the Linux direct boot sequence and show how I taught bhyve about it, also show how QEMU would like to work and why it doesn't quite line up with bhyve's view of the world, and hopefully show you how low-level kernel hacking for any OS can be made as simple as hacking on any other program.
