Responsibility for Enet link state

[Michael ODonnell]

What entity, if any, on an Enet-equipped system is
responsible for seeing to it that the interfaces are
live and operating at (somebody's definition of) the
"proper" speed/duplex/etc?

I'm asking because we have some systems running
RHEL4 with some 10/100 NICs that occasionally wake
up confused such that they're only willing to run at
10baseT-HD when we really need them at 100baseTx-FD.
We know why this is happening (firware diags leave
the NICs scrogged in various ways) but it seems like
every part of the networking infrastructure - from
the e100 driver on up - assumes that as long as the
interface is live it must be running at the desired
speed/duplex settings.

Once we've noticed that the link is not running as
desired it's trivial to use (say) mii-tool to force
the interface to restart autonegotiation and that
always has the desired result, at least in our case.
But I wonder if there isn't some obscure part of the
system config infrastructure which allows the admin
to specify that certain interfaces must run with
certain characteristics and then enforces that.

In other words, after you (or your system startup
scripts) have said something like "ifup eth0" how do
you know that its speed/duplex settings are correct?