> From: mycroft@gnu.ai.mit.edu (Charles M. Hannum)
> Date: 25 Jan 1997 17:43:54 -0500

> "Gordon W. Ross" <gwr@mc.com> writes:
> 
> > 
> > 	arch/alpha/include/espvar.h
> > 	arch/sparc/include/espvar.h
> > 	arch/sun3x/include/espvar.h
> 
> Er, one more thing...
> 
> It seems seriously wrong to have these files.  The only place I see
> where there's a significant performance issue with adding function
> pointers would be espreadregs(), and that really needs to be a
> callback to the MD code anyway.

Could you please explain what is "seriously wrong" about having this
(or any) driver use a <machine/xxxvar.h> file?

How is it different than, say, <machine/param.h>, or any other
case where we allow MI code to use an MD header to let the MD
code have some influence over how things are done.

Gordon