>
> >On 19 Feb 96 at 0:58, James MacKinnon wrote:
> >
> >[about problems with 2940/B]
> >
> >I know that there is a rare problem in 1.3.64 with AHA2940 Rev. B,
> >IBM DPES 31080, tagged queueing, and queuesize=4. From time to time
> >files are read incorrectly. Still. Unfortunately. It seems that a
> >high I/O load makes the problem be more likely.
> >
> >In recent revisions of the driver the problem ist really rare, but
> >still present. Maybe it's the Linux filesystem, I don't know. Does
> >anybody else know? (Ok, silly question).
>
> The latest version of the aic7xxx driver has improved performance. Possibly
> this could uncover a bad motherboard cache or memory problem, though I
> thought you would have seen it before now.
>
> Try disabling the cache on your motherbard, or swapping out SIMMs (if
> you have that luxury). Also check cabling and termination.
>
> [This is mostly for James MacKinnon, not you Ulrich :) ]
>
> Dan EIschen
> deischen@iworks.InterWorks.org
>
I've disabled cache, swapped cache, swapped SIMMS ... problem still
occurs. I have active termination on this card, and it is rock-solid
otherwise with the old driver.
Perhaps I hadn't emphasised this in my previous post:
The problem only manifests itself when the target disk becomes FULL
during the test, so perhaps ENOSPC is being ignored/mis-coded by the new
aic7xxx driver (perhaps in the SCB DMA routines?). Because the IDE drive
is affected, the driver may be over-writing into the I/O space of the IDE
driver.
I'll try some debug on it.
Thanks,
--
James S. MacKinnon Office: P-139 Avahd-Bhatia Physics Lab
Department of Physics Voice : (403) 492-8226
University of Alberta email : Jim.MacKinnon@Phys.UAlberta.CA
Edmonton, Canada T6G 2N5
WWW: http://www.phys.ualberta.ca/~jmack/jmack.html