Re: [virtio-comment] TAB Comments on Virtual I/O Device (VIRTIO) Version 1.0

[Rusty Russell <rusty@au1.ibm.com>]

"Michael S. Tsirkin" <mst@redhat.com> writes:
> On Thu, Feb 13, 2014 at 05:01:24PM +1030, Rusty Russell wrote:
>> Patrick Durusau <patrick@durusau.net> writes:
>> > TAB-556	Conformance targets
>> > 
>> > 	After reading 2.2.1 Legacy Interface, again, I have a suggestion
>> > 	for conformance targets in section 9 (subject to my getting this
>> > 	wrong): 1. Non-Transitional Device - enumerate which sections of
>> > 	normative text I MUST conform to in order to have a
>> > 	Non-Transitional Device. 2. Non-Transitional Driver - enumerate
>> > 	which sections of normative text I MUST conform to in order to
>> > 	have a Non-Transitional Driver. 3. #1 plus ?? = Transitional
>> > 	Device. 4. #2 plus ?? = Transitional Driver with the
>> > 	sub-numbering under 1 - 4 specifying the details of
>> > 	conformance. Yes?
>> I originally wanted to make all the legacy sections non-normative, and
>> thus avoid any MUSTs in there.  It's a bit weird, becuase legacy systems
>> our outside the scope of the current standard, but we really do want to
>> be kind for people in transition.
>> I think Michael had objections to this, though.  So here's my
>> compromise:
>
> BTW apparently "Note" means "non normative" so we probably
> should drop word "Note" that I inserted everywhere.
>
>
>> diff --git a/conformance.tex b/conformance.tex
>> index 0e8b321..d564392 100644
>> --- a/conformance.tex
>> +++ b/conformance.tex
>> @@ -6,3 +6,54 @@
>>  An implementation conforms to this specification if it satisfies
>>  all of the MUST or REQUIRED level requirements defined within
>>  this specification.
>> +
>> +In particular, a device does not need to support legacy drivers, nor
>> +a driver support legacy devices to confirm to this specification,
>> +ie. non-transitional drivers and devices are fine.  (In fact, the only
>> +awareness such non-transitional implementations need that legacy
>> +systems ever existed is specified in \ref{sec:Virtio Transport Options / Virtio Over PCI Bus / PCI-specific Initialization And Device Operation / Device Initialization / Non-transitional Device With Legacy Driver}
>> +for PCI device implementations).
>> +
>> +All requirements for such transitional devices and drivers are
>> +explicitly noted as such, and contained within sections named ``Legacy
>> +Interfaces''.
>
> Hmm this makes it look as if transitional does not
> have to satisfy the rest of specification.
> In fact the reverse is true.
> Maybe add:
> requirements contained in these sections apply to
> transitional devices and drivers only.
> These requirements do not apply to non-transitional devices and
> drivers.

Yes, I'm going to rework the Conformance Clauses over the next few days;
it's the big remaining TODO (and one I'd forgotten about prior to
Partick's feedback).

To do this, my plan is:
1) Revisit all the non-capitalizes may/should/must.
2) Move all notes and examples out to indented "Note:"/"Example:" pararaphs.
3) Put labels on every normative sentence.
4) Write the conformance clause(s).  There will be at least two, one
   for devices and one for drivers.  This might involve reshuffling text
   to group the normative sections for easier references.

I will also explicitly require that any driver MUST interoperate with
any conformant device, and vice versa.  The exception is feature
negotiation failure.

Let's see what it looks like once I'm done...

Cheers,
Rusty.