At 7:53 PM -0500 12/5/04, Michael wrote:
>800kb/s sounds like what I remember getting through mc, any idea why not
>everyone was affected?

I'm actually rather puzzled by this problem. I can definately make it go
away, but I'd like to be able to back out the code changes and make it fail
again, but as I just posted, I'm only partially successful at that. I can't
compile a new GENERIC because I've already mangled so much other code, so I
have to compare it against 2.0RC4 GENERIC (non MP). I'll readily admit that
while I feel comfortable saying this patch is promising, I can't say I've
purely identified the problem.

The apparent randomness of people affected also bothers me. It may have
something to do with how they set up /etc/ifconfig.mc0. It might have to do
with which kernel build they are using. There are clear indications that
the if_mc.c code was being adjusted for the ifnet* interface, but it is
almost like the API was not complete at the time (#ifdef NOTYET) and the
person was waiting for the rest. Clearly the replacement media change and
media select routines in if_mc.c work better and are more consistent than
the am79c950.c code, which basically rejects any media type. I think in the
long run we're better off using them, even if I can't show precisely why
they work and the other ones don't.

Now, if other people start saying that it doesn't fix their mc*, I need to
go over this very closely. I can reproduce it somewhat consistently, which
goes a long way to figuring out if I'm actually changing something.

tim