In message <20020709080050.P12480@dr-evil.shagadelic.org>Jason R Thorpe writes
>Firstly, I should be receiving a few of the new cards shortly, along
>with the docs, from Intel.

Great. I now have a couple of the two-port cards on my desk, waiting.
(Intel has an eval offer on their "server nic" page; its a good deal.)



> > One should, perhaps, consider ditching if_wm.c altogether; 


>No point.  [... performance numbers ]

If the performance bugs have gotten fixed: great. Sincerely.  There's
no reason to ditch the driver framework if that is fixed.  My overall
issue is that -- even compared to if_gx, or the later wx incarnations
-- the *initial* checkin had both abysmal performance, and (sans the
NDA docs) was also much harder to understand, to diagnose bugs, or to
fix them, than the corresponding code in the gx driver.

At least that's how I found it, when I tried to add 82543 to get
head-to-head benchmarking.

Have you looked at how the gx handles the gpio pins for media select
on an 82543, or the other specifics? It works, and it also looks
rather cleaner than the (partial, buggy, broken) code, for the same
chip revs, in the initial wm commit.  Isn't cleaner better?