Subject: Re: netatalk support
To: Bill Studenmund <skippy@macro.stanford.edu>
From: David Brownlee <abs@anim.dreamworks.com>
List: current-users
Date: 09/03/1997 13:13:56
	Would it be reasonable for someone to look at making protocol lkm
	which can be loaded but not unloaded (as a first step), then work
	on making them unloadable?

		David/abs

  -- "Its hard to say it... I hate to say it... but its probably me..." --

On Wed, 3 Sep 1997, Bill Studenmund wrote:

> On Wed, 3 Sep 1997, Tom T. Thai wrote:
> 
> > On Wed, 3 Sep 1997, Bill Studenmund wrote:
> > 
> > > > I couldn't access fugue so I tried to dl the current unpatched beta.
> > 
> > it's strange, but ftping to fogue from netscape timed out, but from shell
> > it works fine..
> > 
> > another thing.. I hate recompiling my kernel and reboot just to add
> > options NETATALK.. what happend to LKM?
> 
> We don't support protocol lkm's right now. Two changes would have to
> happen. First, we'd need to change how protocol interrupts get processed.
> Right now, they are dispatched in M.D. code. Basically the dispath is a
> set of things like:
> 
> #ifdef NETATALK
>         if (netisr & (1 << NETISR_ATALK)) {
>                 netisr &= ~(1 << NETISR_ATALK);
>                 atintr();
>         }
> #endif
> #ifdef NS
>         if (netisr & (1 << NETISR_NS)) {
>                 netisr &= ~(1 << NETISR_NS);
>                 nsintr();
>         }
> #endif
> 
> We need to generalize that to look at slots in a jump table. We also need
> to generalize the input and output routines in if_ethersubr.c to handle
> added-on-the-fly routines.
> 
> Then, we need to re-vamp things so that we can remove a protocol cleanly.
> I'm working on changes to permit removing interfaces, and it's turned into
> a lot of work, because the current networking code just isn't make with
> disapearing ifnet's in mind. I think we'd run into the same thing with
> protocols (though probably not as bad).
> 
> I'm not saying we shouldn't do this, it's just we have a bit to do first.
> 
> Take care,
> 
> Bill
>