Adaptec 7xxx Support

["Adam Williams" <awilliam@xxxxxxxxxxxxx>]

>
> > P.S.  I think I've fixed the "Die-ing in it's sleep server."  I browsed the
web
> > for several hours and ran across complaints about Linux with a Adaptec 2940
> > SCSI Card (actually a aic7880).  I got it to be stable using a module
option,
> > linux kernel 2.0.31,  and disable-ing "Wide Negotiation" in the card's
BIOS.
> >  Thanks to Bruce and R.G.B. for your help.
>
> Do you have the info you found on this problem?  I'd like to look at
> it since I have that same SCSI card in my PC at work.  Thanks.
>

First I discovered that I could "reliably" crash my machine by using dd to raw
read the partitions and write them out to files near the end of the disk.  I
was truly grabbing at straws but after three nearly immediate crashes I figured
that the hard disk or controller were at fault.  Since the drive is an IBM
Ultrastor,  of which I have nearly a dozen operating flawlessly in my database
server,  I turned by attention to the SCSI controller.

I did a search from Alta-Vista on the Usenet as well as looking a a brand new
copy of the Linux SCSI-HOWTO (Dated Dec. 21).   This is a direct quote from one
of the postings "Defective 2940 drivers (AIC7xxx) are pandemic on Linux.".  The
errors reported varied from hung systems, to random floods of SCSI errors, to
signal 11's issued when compiling the kernel (schulze@LPL.Arizona.EDU).

Fix number one handles the addressing of the drives,  which can become confused
after a period of time.  This if applied by passing the mentioned module option
to the kernel at boot time.
My lilo.conf snippet:

"  image=/boot/vmlinuz-2.0.31
"         label=linux
"         root=/dev/sda1
"         initrd=/boot/initrd-2.0.31.img
"         append="aic7xxx=extended"
"         read-only

The second fix is actually mentioned in the Red Hat 5.0 manual,  and that is
too disable "Wide Negotiation"  as Adaptec is still adjusting that feature at
will,  and not releasing the information to developers.  You can do this
through the config utility accessed by pressing Ctrl-A prior to boot.  Of
course I only found it in the manual after digging through Usenet for several
hours.

It's possible that either one of these fixes,  in and of itself,  might have
fixed the problem.  But given that this machine is a NIS and DHCP server that
should have been online a week ago I don't have the opportunity to experiment.

One other message stated that the 2.1.59 supported the Adaptec card with no
jerking around,  and that that kernel had been stable for him 24/7.
 (szb50@amdahl.com).  But since my current fixes worked (at least so far) and I
expect 2.1.x to go 2.2.x (production) pretty soon I haven't tried this either.

As a piece of trivia,  several of the posts stated that they had made this card
stable on multi-processor boxes as well, using the lilo parameter.   From what
I read the card seems to be less stable the more powerful the machine.  In one
case the card worked fine with a 100MHz pentium but blew chuncks at 120MHz.