On Thu, 01 Dec 2011 09:55:30 +0100, Paolo Bonzini [off-list ref] wrote:

On 12/01/2011 04:14 AM, Rusty Russell wrote:

quoted

I'd prefer to see the spec only cover things
which are implemented and tested, otherwise the risk of a flaw in the
spec is really high in my experience.

In general I agree, and I did that for virtio-specific things such as 
the eventq and the configuration space.  This is also why I don't want 
to add untested controlq requests that people suggested.

However, there's tension between this and providing a complete SCSI 
transport.  SCSI is roughly defined as a set of RPC interfaces ("Send 
command", "Abort task", etc.) and transports provide the RPC protocol. 
The SCSI command set changes relatively often, but the RPC interfaces 
are pretty stable.  This stability limits the risk, and having a mapping 
for all interfaces also makes future changes less likely.

OK, I expect that someone reading the spec will be SCSI-familiar more
than virtio-familiar, so it's more important that they be comfortable,
than that I be comfortable.

Thanks!
Rusty.