Subject: Re: spontaneous coredumps?
To: Todd Whitesel <toddpw@best.com>
From: Hauke Fath <hauke@Espresso.Rhein-Neckar.DE>
List: port-mac68k
Date: 11/13/1999 16:35:39
At 13:59 Uhr +0100 13.11.1999, Todd Whitesel wrote:
>I have been using the "GENERIC" kernel without any modifications.

[...]

>While building a kernel I ran into the % transition in the assembly code,
>and built/installed gas. After that the kernel build proceeded reasonably
>far along before dying; interestingly, it didn't die of a core dump event,
>but rather due to misspellings in .depend ("don't know how to make xxxx").
>Two of them:
>	./../../../sys/types.h
>	../../../../syss/dirent.h
>
>Now the twilight zone klaxon is _really_ going off. Dropped/duplicated bytes
>is a very scary kind of failure.

Sigh. You fell prey to ncrscsi... I've got a Quantum LPS105S where ncrscsi
drops one byte every 16k during writes.

Try GENERIC.SBC; the vanilla GENERIC kernel config has the ncrscsi driver.
Both have their share of problems (that's why they are both available); the
recommended strategy is to start off with ncrscsi (which is said to be a
bit faster and does disconnect-reconnect out of the box) and switch to an
sbc kernel.

>BTW, as I recall the 68k Macintoshes never really supported SCSI Arbitration,
>ever. This was frequently declared to be a software problem, in the form of
>non-existent support for it by the Mac Toolbox SCSI Manager.  But perhaps it
>is related to a real hardware issue, since the hardware was never tested
>against an arbitrating SCSI driver?

Apple's SCSI-Manager 4.3 does not support the 53c80 and I suspect they know
why -- although they used the damned chip in all the Powerbooks even after
they switched the desktop models to 53c9x.

The esp driver is better in that respect, but not much: Two Quantum Atlas
III drives that would not even be newfs'ed on both a IIci (sbc) and a Q700
(esp) are doing just fine on a Sun IPX (also esp).

From what I see NetBSD/mac68k suffers from a manpower problem wrt. SCSI.
The two people who have been working on mac68k scsi are Allen Briggs and
Scott Reynolds, and they both have more than enough other work at their
hands...
>
>Is it possible to run our scsi drivers without disconnect/reselect, and has
>anyone tried this on mac68k to see if it improves anything?

The sbc driver should fix your problems. Note that there are configure
flags to [en,dis}able disconnect-reconnect and interrupt-driven transfers.

	hauke


--
"It's never straight up and down"     (DEVO)