cmcmanis@mcmanis.com said:
> My question is what is the root of this strange design? I'm probably
> going  to write a small and verifiable set of RPC functions into libsa
> that are  both portable and readable (I've worked on RPC in the past)
> and they should  do things in a somewhat more predictable way.

As people wrote, code size is an issue. Memory fragmentation too.

To help you debug your problem: You can try to test the libsa
stuff in userland. See the code in arch/i386/stand/lib/test -
there is glue code to use a bpf as network interface, and if
your platform allows, you can also directly access some hardware
(which shouldn't be controlled by a kernel driver:-).
This used to work a while ago. It's not unlikely that it got
slightly broken by recent changes, but should be easy to fix.

best regards
Matthias