Jonathan Stone wrote:
> 
> on  Fri, 04 Jul 97 21:36:04 -0700, "Chris G. Demetriou"
> <cgd@pa.dec.com>  writes:
> 
> [very solid evidence of compiler bug snipped]
> 
> I'd been compiling -current kernels for i386 with -O6
> -feliminate-frame-pointer between approximately 8 months.  I stopped
> and went back to -O2 sometime in about May, after installing the
> 'upgrade' to gcc 2.7.2.2, because it caused kernel crashes.  I _think_
> all the machines that crased had 3c595s.
> 
> I'd guess the two are related; maybe it could help someone pinpoint
> the bug a little more easily.


Another data point - about a month ago, I started experiencing
nightly system panics.  It turns out that the system would panic 
in the vm subsystem during heavy disk (NCR 53c810) accesses.
Note that I don't have any NICs in my system.

At the time, I was compiling my NetBSD-current/i386 kernel with
the Pentium enhanced version of gcc 2.7.2, at optimization level 3.
So I switched to the shipping gcc, still at -O3, and the situation
didn't improve.  Backing down to -O2 (pgcc and gcc) has since
solved the problem.

Since I haven't changed pgcc within the past 6 months, I don't think
the problem is related to changes to the compiler.  I believe the 
problem is either related to the series of vm changes that occured
1-2 months ago, or the said changes revealed a bug in the compiler
(2.7.2 and 2.7.2.2).  I can provide a more details bug report, if
anyone thinks it will be helpful...


Jeff Thieleke