Re: Adaptacrap 1542C vs 1520B
Hi,
> On Wed, 26 Jul 2000, Ron Farrer wrote:
>
> > What would work better in an Alpha; an Adaptacrap 1542C or 1520B?
>
> Don't think either work. I tried an Adaptec 1542C in the past and it
> never worked due to horrid dma/memory mapping and, from what I understand,
> it's not fixable.
The problem is, that DMA transfers can't go beyond 16Mb on ISA native.
However, I have seen drivers using bounce-buffers below 16mb, moving data
to/from the target memory-space.
A ( guick ) glance through Linux' scsi drivers revealed, that the
ISA Buslogic cards seem to make use of this method.
However, looking at the aha1542.[hc] revealed no clear picture ( to me ).
It seems, that *any* transfer beyond 16Mb results is a panic() call.
( see macro BAD_DMA in aha1542.c ).
Actually, I found it hard to believe, that this is actually the case. This
problem has been fixed in numerous x86 *NIX implementations ( at least
Solaris and SCo did, to my knowledge ).
Also, I assumed, that modern PCI-ISA bridges woudl fix this problem
"automagically". However looking at the generic isa-dma code for Alphas,
there appears to be a limit at 48Mb.
> I haven't tried the 1520B, but I'm assuming it's more of the same. In my
> experience, it's much better to get a PCI SCSI adapter (I've got a Tekram
> 390F...only like $70).
The 1520 uses polled I/O, not DMA. I've never seen one - it was said to
be an awful slow dog at the time it was introduced. But I assume, it *could*
work.
Also, if you ( Ron Farrer ) has the option of choosing, I would probably go
for a BusLogic. At least the driver uses bounce-buffers.
Anyway, it's very late now ( in Germany ) and i'll finish now.
Just my $.02
Sincerely
Thomas Weyergraf
> --
> To UNSUBSCRIBE, email to debian-alpha-request@lists.debian.org
> with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
>
--
Thomas Weyergraf kirk@colinet.de
My Favorite IA64 Opcode-guess ( see arch/ia64/lib/memset.S )
"br.ret.spnt.few" - got back from getting beer, did not spend a lot.
Reply to: