The more I look at the preload way of doing things, the more I see this as
something of a lose.
What we want to do is to load the fw module if and only if there is h/w for
it, and then unload it when we're done. We want to do this w/o user
intervention, else why bother (as we could have gotten them to run config
compiling in f/w if the want to support this card).
What we have the tools for is to get the loader(8) to load it for you, if you
request it (i.e., by modifying /boot/loader.conf to specifically *add* it, or
by modifying /boot/defaults/loader.conf to make default loaded (leaving to the
user to edit /boot/loader.conf to disable the load), plus (I haven't done this
step yet) some goop to run around and hand bind it from isp_pci so you can
either get at the data directly, or (which I would prefer) get at the function
that isp_pci will call which will pass back a pointer to f/w.
But there's nothing there to unload this module if you *don't* have the h/w
(i.e., all probes for isp fail), unless I stick the PCI lookup info into this
module (and it unloads itself... can it? Is there a mechanism where we can
mark a container for unloading later so it can be scavenged?).
So, what we end up with is more text code, a bunch of new places to touch
(like defaults/loader.conf) and nothing to reduce the bloat (which was the
whole point of this exercise) if you don't have the h/w.
*groan*
-matt
To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-hackers" in the body of the message
