On Wed, Apr 10, 2019 at 08:15:32PM +0100, Stefan Hajnoczi wrote: > On Wed, Apr 10, 2019 at 4:45 PM Yoni Bettan <ybet...@redhat.com> wrote: > > On 4/9/19 4:17 PM, Stefan Hajnoczi wrote: > > > On Mon, Apr 01, 2019 at 02:18:43PM +0300, Yoni Bettan wrote: > > The final purpose is to have: > > > > 1. device specification > > > > 2. device implementation > > > > 3. device driver > > > > 4. blog > > > > maybe I should have written it at the beginning, this is not the entire > > project but it is its start. > > The way I'd design VIRTIO devices without prior knowledge is: > > 1. Learn the VIRTIO device model. Understand the concepts in VIRTIO. > <-- this is hard today, there's not much good documentation
Best doc is still probably Rusty's whitepaper. It only covers 0.X spec so somewhat outdated but it does explain the concepts I think. > 2. Design an initial version of the device spec. Mostly configuration > layout, virtqueues, and request structs. Not much text is necessary > at this point, but it's critical for thinking through features before > implementation. > > 3. Implement guest driver and device emulation. > > 4. Iterate on spec and implementation until it's functionally complete. > > 5. Submit the spec to the VIRTIO Technical Committee. > > 6. Submit driver and device emulation patches. They can be merged > when the spec is approved/close to approved. > > Are you jumping to #3? This is likely to lead to poor quality > implementations and specs because the fundamental VIRTIO concepts > weren't understood. > > If the point is to educate others and/or do it "the right way", then I > would really avoid hacking around without first doing the other steps. > > Stefan