On Sat, Jul 21, 2001 at 10:42:23AM +0200, Ignatios Souvatzis wrote:
> On Tue, Jul 10, 2001 at 03:01:54PM -0500, Rob Healey wrote:
> 
> > [...] warnings being treated as errors
> > ../../../../netinet/ip_gre.c: In function `gre_input2':
> > ../../../../netinet/ip_gre.c:194: warning: implicit declaration of function `setsoftnet'
> > 
[...]
> 
> if_gre.c both as of 1.5.1 or of current DOES include <machine/cpu.h>.

Sorry ... I looked at the wrong file, obviously.  So, my opinion is this:

ip_gre.c tries to get the splfoo() definitions automagically via other
header files. I don't know why this does work on some architectures; on
Amiga (and as far as I can tell, at least on Atari) it fails.

Yes, ip_gre.c should include machine/cpu.h, as if_gre.c does already
(or, maybe, intr.h). If there are no objections, I'll change this on
-current and ask for pullup.

Rob: please test whether machine/intr.h is enough.

Actually, maybe netisr.h should handle this internally. In this case, netisr.h
needs to pull in at least machine/intr.h to the very least...

What's our policy?

[discussion moved to tech-kern]

Regards,
	-is