Subject: SOLVED: random cc1 SEGV's with 1.4.1/1.4.2_ALPHA
To: None <port-i386@netbsd.org>
From: Todd Whitesel <toddpw@best.com>
List: port-i386
Date: 01/28/2000 00:48:21
Underclocking the CPU & motherboard (relative to their documented abilities,
at least) fixed the problems.

Details, for those fascinated by weird hardware behaviors:

This motherboard is an Iwill P6NS, originally purchased to run a 150mhz PPro
(cheap at the time) so I could upgrade it to a 300mhz one when that came out,
except of course it never came out -- Intel stopped the PPro at 200mhz.

It is a very nice motherboard with on-board AHA2940UW, and I didn't want to
replace it yet. So when I saw Intel offering the "Pentium II OverDrive" PPro
upgrade last year, I waffled for a week and then paid the "soak em" price
(even after the mail order discount).

The P2 OverDrive can run at 300mhz while impersonating a 150/180 Mhz PPro,
or run at 333mhz while impersonating a 166/200 Mhz PPro. As far as I can
tell this works fine; Intel quality blah blah blah.

At the 150/180 processor settings, the motherboard "runs at" 50mhz according
to the BIOS messages; at the 166/200 processor settings it "runs at" 75mhz.

Experiments show that the 50mhz settings work totally perfectly. I can run
multiple compiles on repeat (via shell scripts) for hours and hours without
any kind of failure. But at the 75mhz settings, the same test load takes 25
minutes to warm up and then segv's occur every few minutes after that.

This is really sad, because the extra 50% in the memory system really helps.
I suspect that either Fry's palmed me 70ns chips when I bought 60ns, or that
the memory system doesn't totally work at the 75mhz setting but the original
testing methodology was so PC-based it could never have caught this problem.

I could test this by using BIOS settings to slow down the configuration that
doesn't work until it starts working. However, the only BIOS settings slower
than 75mhz-full-speed are nearly equivalent to 50mhz-full-speed, and that
basically negates the performance advantage (although it would let me run
the CPU 10% faster).

This is my designated "stable" machine and I can certainly live without an
extra 10% CPU. And the machine is still fast enough for what I do on it.

Strategic withdrawal time.

Todd Whitesel
toddpw @ best.com