sowmini.varadhan at sun.com wrote:
> On (07/18/07 18:50), Garrett D'Amore wrote:
>
>> Actually this is one of those cases where I think the existing interface is
>> so busted that nobody will complain too loudly if we just abandon the old
>> interfaces. I will argue the case this way if it is so presented at PSARC.
>>
>> -- Garrett
>>
>>
>
> that is true, but since we won't be able to convert every driver overnight,
> we'd end up with a situation where some drivers- the ones not yet
> switched to Brussels- do the "busted" interfaces, while others do the
> "new, improved" implementation. We could either just leave the "busted"
> interface undocumented while we convert them all (the average reader
> probably won't get much useful info out of the existing truth table,
> anyway), or we could leave the existing documentation for whatever it
> is worth and also add in the new definitions.
>
No, the old documentation is busted... because few if any drivers
actually *follow* that truth table. Apparently its just as hard for
driver developers to make sense of as it is for end uses. :-/
So, gut the table, leave the old drivers unconverted for now, and
document the new way forward.
By the way, we *really* need to have a "summit" of some kind to better
codify what the various kstats drivers use are supposed to mean. There
are problems here with inconsistent ideas of what is a jabber, a macrcv
error, etc. (Do opackets correspond to actually transmitted packets,
or just transmit attempts. Etc.)
The kstats in Nemo are almost as much of a mess as they were in previous
DLPI drivers; they could really use some attention. I'd like Brussels
to be the time to do this... since folks are going to be in that portion
of code converting for Brussels, it makes sense to sanity check that
they are doing the statistics accounting properly too. But first we
need to have a better definition of just what that means.
-- Garrett
> --Sowmini
>
>
