>VLB and SCSI are the LINUX' best friends.
>I've got some trouble by implementation of LINUX on
>a 486DX2/66 with VLB and AIC 6360 as SCSI-hostadapter.
>The HDD is a MAXTOR 7345S and there is a CD-ROM TOSHIBA
>First we generated a new kernel with -DSKIP_BIOSTEST.
>It found the adapter and the HDD with no problems.
>But at irregular points errors like
>aha152x: P_CMD: 1(0) bytes left in FIFO, resetting
>appeared and the system halted or booted uncompletely.
>Now I would be happy, if someone will help to answer
>one of the following questions:
>- Does a explaination to the above errors exist ?
Yes. There is a timing problem in the Adaptec 1520 driver which
is in the distribution kernel which manifests itself when used
with VLB cards, and may also be a problem with FIFO under/overflows.
A fix from Juergen follows for the first problem, and we still
haven't tracked down the roots of the second.
Quote:>- Are there better adapters (for instance the ULTRA 34F or FA)
> for SCSI/VLB/Linux ?
The Buslogic 445S - very fast, Adaptec 1542 compatable so support
is solid under Linux and other systems.
Quote:>- Is the combination SCSI/VLB/Linux impossible in general ?
Quote:>- It is better to work with a SCSI-ISA-adapter on a
> VLB-board by Linux? (How much is the influence
> on the perfomance?)
Throughput is usually unaffected by the bus, CPU usage
is determined not by bustype but by PIO/FIFO buffered/bus mastering,
and percentage of bus bandwidth used (ie, impact on other bus
hogs like video performance) is determined by bus type.
Quote:>IS THERE ANYWHERE A COMBINATION VL/SCSI/LINUX, WHICH WORKS?
Linux + Buslogic 445S + cluster patches + adaptec 1540 driver
works extremely well for many people.
I've appended Jurgen's fix, it won't hurt anything to give it
a try -
Subject: aha152x patches for VLB boards
Date: Tue, 1 Feb 1994 09:57:18 -0100 (GMT-1:00)
X-Mailer: ELM [version 2.4 PL17]
Hi Folks !
We've found the problem with the aha152x code on vlb bus machines. It
seems that a insw is to fast for 6360 on vlb, so that it can't notice
every read attempt and so the driver thinks it has read more data than
the 6360 has supplied. I've substituted the insw with a inw loop and
with this change it seems to work.
Patches (btw 0.102 is the original source in p14x):
--- 0.102 1994/01/31 20:44:12
/* aha152x.c -- Adaptec AHA-152x driver
- * Copyright 1993 Juergen E. Fischer
+ * Copyright 1993, 1994 Juergen E. Fischer
#undef DEBUG_QUEUES /* debug reselection */
/* recently used for debugging */
+ int words;
+ unsigned short *ptr;
CLRBITS(DMACNTRL0, _8BIT );
data_count >>= 1; /* Number of words */
insw( DATAPORT, current_SC->SCp.ptr, data_count );
+ ptr = (unsigned short *) current_SC->SCp.ptr;
+ words = data_count;
+ *ptr++ = inw( DATAPORT );
/* show what comes with the last transfer */
Please try out these patches by recompiling with DEBUG enabled (VLB will be