>> We can do that because we use 64-bit off_t's, as opposed to the
>> 32-bit off_t (well, long) that SunOS used.

> The disadvantage, however, is that when compiling stuff on my Sun3/60
> under NetBSD, the 64-bit off_t's cause lots of troubles.

Only if you have code that doesn't use the off_t type.  If it insists
on using long and thinking it knows a priori that off_t is long (or
more to the point, that the second argument, and return value, for
lseek, should be long), yes, you will have trouble.

You'll also have trouble with code from PDP-11 days that "knows" that
short and int are really the same.  The only essential difference I see
is that nowadays, it's much more common to "know" that off_t==long than
it is to "know" that short==int.

> How could I make si_options 1 permanent into the kernel?

Edit sys/arch/sun3/dev/ncr_si.c and change the initialization of
si_options.  I'm actually considering adding code to make it a kernel
config option, so that my config file can say something like
	options		SI_OPTIONS="SI_ENABLE_DMA|SI_DMA_INTR"

					der Mouse

			    mouse@collatz.mcrcim.mcgill.edu