[1375] in linux-scsi channel archive
Re: NCR53C8XX (BSD ported) 16e for 2.0.28, Hard error.
daemon@ATHENA.MIT.EDU (Gerard Roudier)
Sat Feb 1 05:52:56 1997
Date: Sat, 1 Feb 1997 11:49:03 +0000 (GMT)
From: Gerard Roudier <groudier@club-internet.fr>
To: linux-scsi@vger.rutgers.edu, ncr53c810@colorado.edu
cc: "Jake (not the snake) Messinger" <jake@ams.com>,
Stefan Esser <se@freebsd.org>,
Krzysztof Adamski <kadamski@netsurf.net>,
=?iso-8859-1?Q?J=FCrgen_Sahlberg?= <sahlberg@fe.uni-bremen.de>,
ncr53c810@colorado.edu
In-Reply-To: <19970131214926.PJ17500@x14.mi.uni-koeln.de>
On Fri, 31 Jan 1997, Stefan Esser wrote:
> On Jan 31, jake@ams.com (Jake (not the snake) Messinger) wrote:
> > On Fri, 31 Jan 1997, Stefan Esser wrote:
> > > Didn't the SC200 switch over to using the 810A,
> > > like Symbios did for their Sym8100SA cards ?
> >
> > Whats the difference?
>
> Many details, but the most important is
> support of burst op-code fetches, IMHO.
>
> The 810A just makes better use of your
> PCI bus. I don't have the docs here, but
> AFAIK, the 810 did not support read-line
> and read-multiple PCI accesses (which do
> reduce the cache snoop-overhead on the
> CPU) or write-and-invalidate (allowing
> stale cache contents to be flushed
> instead of being written back by the CPU
> and then immediately overwritten by the
> SCSI chip ...).
>
> The read-multiple command seems to cause
> problems (at least in combination with
> some other conditions), and this was the
> start of the thread, IIRC :)
Disabling Read Multiple did not fix the problem.
I have received a report that shows the SDMS BIOS set DMODE to 0x8e
for 810A board, so it enables all features you describe above. I do'nt
know if "write and invalidate" is set by SDMS BIOS for such chips.
Assuming that all scsi boards have been heavily tested under M$$
software, I am not sure about the current diagnostic for James's
problem.
Unfortunately, we do not receive success reports with accurate hardware
description.
Being given that all recent 810 based boards use 810A chip, I think we
should wait for possible other reports about similar problem.
For now, people who are using NCR53C8XX 1.16e or 1.17 with 810A
controller, and have problems, should try to boot their system
with burst 8 and _no_ PCI dma feature enabled by the driver
and let me (us) know.
Boot command line: ncr53c8xx=burst:3,specf:n
People who have had success with the driver 1.16e/1.17 and 810A chip
can mail me directly (no cc needed to lists) their hardware description.
That may help.
Gerard.
