Network device driver KPI/ABI and TOE

[Andre Oppermann]

Peter Jeremy wrote:
> On Sun, Jan 06, 2008 at 06:37:19PM +0000, Robert Watson wrote:
>> What I'm more concerned about is the new exposure of internal data 
>> structures and algorithms, and a resulting freeze of those data structures 
>> and algorithms if we were to apply our current ABI/PI policy to the TOE 
>> interfaces.
> 
> Whilst I doubt TOE will directly affect me in the short term, I would be
> disappointed if general TCP improvements could not be MFCd because it
> would change the TOE ABI.
> 
> I believe that TOE is fairly new and not completely mature feature.
> Is it possible that further experience with TOE may also lead to
> changes in the interfaces between TOE and the rest of the kernel,
> irrespective of the kernel innards?

Certainly.

I agree with Robert that we should not guarantee a stable TOE KPI/ABI
yet.  TOE is a relatively young technology and so far we've only seen
one hardware for it together with its assumptions and implementation
issues.  It is also way more complex than simpler features than checksum
offloading or segmentation offloading.  IMHO we should not make it fully
stable unless we've gained more experience with it and also a second or
third hardware making use of it with perhaps slightly differing
assumptions and requirements.  OTOH there shouldn't be any deliberate
breakage of TOE without a good justification for it.

-- 
Andre