On Thu, Jun 22, 2000 at 02:24:02PM +0200, Thomas Niesel wrote:
> Dear all.
> I wonder if you could help me with my controler ?
> The hardware in details:
> Tekram DC800B, ISA-cache-mirroring-controler with onboard-floppy-port, up to
> 16MB Cache-memory (4x4MB from the good old time).
> Chipset: NCR55CF94-2 and an 80286-20Mhz-cpu (no joke !)
> Others: io0x330, irq11, dma5, bios from PC shadowed to 0xdc000 and the bios
> from the controler also shadowed on 0xdc000 as the recomendaiton of the
> manual, set max.DO$ capacity to 1GB. (the disks are 2GB each !)
Nice piece of legacy hardware. Does it only do mirroring (RAID1) or also
more funny things like RAID5?
> ------------from the manual-------------------------
> MAX DOS disk Capacity: 1G/1.5G/2G/3G/4.5G.
> 1G is the default option and compatible with AHA 1540/1542 series, that
> means you can directly port the drive with AHA 1540/1542 series format to
> this ADAPTER.
If you move SCSI harddisks from one host adapter to another one, there
is normally just one thing that can go wrong: Your SCSI Bios might be to
stupid to accept the geometry of the partition table. Adaptec's BIOSes are
known to do that ...
> For operating systems other than DOS and NetWare, you must
> choose the 1G option. Also, to activate options other than 1G, i.e.,
> 1.5G/2G/3G/4.5G, you have to enable the DOS 5 Support Enable/Disable in the
> Adapter Firmware Options first.
As far as I can see, this only affects operating systems (like DOS) that
access the drive via INT13h from the BIOS.
> The following is the parameter translation
> infor-mation corresponding to each option (512B/sector):
> 1G = 64 heads and 32 sectors, maximum 1GB
> 1.5G = 64 heads and 48 sectors, maximum 1.5GB
> 2G = 128 heads and 32 sectors, maximum 2GB
> 3G = 128 heads and 48 sectors, maximum 3GB
> 4.5G = 192 heads and 48 sectors, maximum 4.5GB
> -------------cut--------------------------------
> Runs on f.e. potato-frozen as an aha1542, autodetect but only with
> 500-700kb/s (cp 300MB-file from cd to disk, between the disks and stoped the
> time).
It really surprises me that the aha1542 adapter can drive such a beast. So
the hardware emulates the aha154x's chipset? Or did they actually put some
aha154x chipset on the ISA side of the controller?
600kB/s is very slow, and I believe the (original) aha154x can do more than
3MB/s.
> The controler(10MB), the disks(20MB) and the cdrom(12x) can do much
> faster.
I guess you should be able to transfer up to 10MB/s on the SCSI bus, and
much more from the card's cache. However, ISA DMA does only allow a few MB
per second.
> I also tried the real size with DO$5 support, same result.
> So I did all the things from the manual: Avoid irq9 cause of sharing
> with irq2, leave it on io0x330 for autodetect, leave it on dam5, disable DO$5
> support. I tryed also different parameters as the listed ones above. Finaly I
> installed DO$6.22 and the driver therefor, and there it showed the power.
> (copy same 300MB-file) gives me a throuput above 3MB/s! An old bench-prog
> sayed: write 3,8 / read 8,8MB/s.
The 8 MB/s surprises me, as that's more than ISA DMA can do, as far as I
know. Maybe you can PIO the data over the ISA bus with that speed, wasting
your CPU cycles though.
> Not bad for the old 386' PC. There is no
> information in linux-docs about this controler. I think the aha1542-module is
> the right on, but I think there must be a different way to talk to the
> controler wich claims to be fully adaptec1540/42 compatible. (io and
> commands)
Maybe you can talk to the aha154x author about this?
Maybe he has (or maybe Tekram has) docu about your controller, so you can
work out how to make it as fast as possible.
> Here is what I think:
> 1st. maybe the translation for the geometry
> of the disks is different from the disk through the controler ?
Don't think so.
> 2nd. the formated block-size on the disk is not the right one ?
Don't think so.
> 3rd. the controler is too intelligent and gets confused
> by the way the driver talks to him ?
Sounds like the most probably reason for your trouble.
> What do you think ?
> Is it helpfull to debug from DO$ on io330 and send the result ?
Not to me ... Maybe to the aha154x authors?
Look into the sources ...
BTW; the settings concerning DOG5 support should not affect Linux, as it
does not use the int13h interface.
Regards,
--
Kurt Garloff <[EMAIL PROTECTED]> Eindhoven, NL
GPG key: See mail header, key servers Linux kernel development
SuSE GmbH, Nuernberg, FRG SCSI, Security
PGP signature
