Subject: Re: Urgent request for help
To: Monroe Williams <monroe@criticalpath.com>
From: Donald Lee <donlee_ppc@icompute.com>
List: port-macppc
Date: 09/12/2001 22:52:35
It is possible that you are having trouble with cables and/or
termination settings, but my bet is that the driver has bugs and/or
the drives and the card driver do not get along well.

As long as you have a proper twisted pair cable (not flat ribbon)
and a proper terminator (lvd!), and don't have some drive terminator
set on in the middle of the bus, I'd say the cables are probly OK.

I have tried a 39160 card in my PM 7600 and have seen the
"someone reset channel A" problem.  It repeats the message
continually and refuses to boot.

I have personal experience with the Adaptec chips, and know
that they are complex and have a number of "undocumented restrictions".
It amazes me that the NetBSD team has built a functional driver for
these cards (2940, 2940U2W and 39160 families).  This is especially
impressive considering that from what I've seen, the work has
been done without benefit of a SCSI analyzer.

Let me add a couple of data points, FWIW.
There are four chip families for the common cards, and each needs
slightly different microcode for the "sequencer" on the chip.
I rememebr the families, but not the chip names or numbers.

The 2930 card (8-bit ultra) is based on the "Talon" chip.
The 2940UW card (16-bit ultra) is based on the "????" chip (I forget)
The 2940U2W, 2940U2B (16-bit Ultra-2) is based on the katana(?) chip
The 29160, 29160N and 39160 (16-but Ultra160 LVD) are based on the
	Trident 1/2 chip(s)

Of these four basic families, the earliest, slowest ones are the
most simple and straightforward inside.  The 2940UW is also the most
common one "out there".  I would expect the drivers, therefore to
be the most stable.

The most recent ones, esp. the 39160 (dual channel 160) are pretty
complex, and relatively new.

In particular, the machinery on the chip to do the termination and
read the NVRAM is tricky and can get you in trouble if you're not
careful.

Barring better advice from someone on the list (author of drivers??)
I'd go with one of the older cards.

-dgl-


>I'm running NetBSD-1.5.1-macppc on a server where availability is rather
>important.  It's currently running on a 9600, and over the last couple of
>years I have seen uptimes in excess of 140 days.
>
>I recently upgraded the storage subsystem by adding an Adaptec 29160 and
>paired 18G IBM LVD drives running mirrored under softraid.  (I was
>previously using the internal SCSI bus for all drives, and was looking for
>better I/O throughput.)  Since the upgrade, I've been having nightmarish
>SCSI problems.  I've tried several different cables, even a different SCSI
>card (an Adaptec 2940U2W), and it's still giving me fits.  The problems
>manifest in several ways, including:
>
>/netbsd: Timedout SCB handled by another timeout
>
>and
>
>/netbsd: sd1(ahc0:9:0): SCB 16 - timed out in Data-out phase, SEQADDR ==
>0x5d
>/netbsd: SCSIRATE == 0x95
>/netbsd: sd1(ahc0:9:0): BDR message in message buffer
>
> [snip]
>
>Please cc: monroe@criticalpath.com with any responses.  I'm on the
>port-macppc list with my home address, but I really need to get responses as
>soon as possible.
>
>Thanks,
>-- monroe
>------------------------------------------------------------------------
>Monroe Williams                                  monroe@criticalpath.com