Subject: Re: Instruction Access Exception on SS20
To: Jon Buller <jon@bullers.net>
From: grant beattie <grant@netbsd.org>
List: port-sparc
Date: 09/25/2002 23:59:05
On Tue, Sep 24, 2002 at 09:53:23PM -0700, Jon Buller wrote:
> >> What's the size of the various kernels? I've recently hit a problem
> >> where the kernel size of GENERIC is fine, but adding IPSEC makes it
> >> unbootable and I have to remove other things.
> >
> >I'll have to check and get back to you. Your comment confirms a suspicion I
> >had though.
> 
> When I got a Instruction Access Exception error instead of a Data
> Exception (or whatever it is when you use -msupersparc) I noticed
> that the kernel was huge, and in that instance, removing the
> -msupersparc reduced the size of the kernel significantly.

The error from the PROM seems to vary (depending on the size of the
kernel?)

>> NetBSD/sparc Secondary Boot, Revision 1.12                                  
>> (grant@mofo, Thu Aug 22 02:33:29 EST 2002)                                  
Booting netbsd.cross                                                           
3042448+101164+288616 [190368Memory Address not Aligned                        

>> NetBSD/sparc Secondary Boot, Revision 1.12                                  
>> (grant@mofo, Thu Aug 22 02:33:29 EST 2002)                                  
Booting netbsd.cross                                                           
2950144+100860+252848 [187952+143622Instruction Access Exception               

> My assumption is that using -msupersparc makes the kernel faster,
> but also bloats it to near the breakage point, and small things
> can push it over the edge.  (All my kernels have IPSEC in them,
> but I clear out all the sun4 and sun4c cruft from my config files,
> not to mention the VME bus stuff and antique disk controllers,
> etc.)

Yes, that is consistent with my findings. I submitted a PR initially
thinking this was a cross compile problem with IPSEC. Further
investigation revealed it is a kernel size problem, cross compilation
does not seem to cause issues. (kern/18138, but probably should be
port-sparc).

g.