On Mon, 21 Oct 1996, Michael K. Johnson wrote:
> I think that this bug is fairly generic. Here at Red Hat, we've had
> reports that moving from modularized low-level SCSI drivers to drivers
> compiled into the kernel has fixed the exact same problem (the card
> being found, but no devices being found on the bus) for people with
> aha174x's and DPT's using the eata_dma driver. I'm currently looking
There seem to be cases in which even this doesn't help.
We have reports from people saying that "old" kernels (up to 2.0.12) work
very well on their systems, but starting with 2.0.13 the aic7xxx driver
doesn't work anymore.
With the DPT there's another problem. I never managed to use this driver
as a module. The module loads fine and detects all devices properly but
then it gives a kernel panic saying
"Attempt to allocate device channel 0, target 0, lun 0"
I do not think that it's a problem of this specific driver because one
person reported to me the same problem with the advansys driver (I cannot
test this, because we don't have such a card).
I can reproduce the problem with the DPT up to 2.0.18 (as the card is in
one of our main servers, I couldn't try it with newer kernels until now).
> into ncr53c8xx (the driver that appeared not to work is Drew's, not
I cannot confirm this. What's the problem? The ncr53c7,8xx works without
any problem when loaded as module. At least until now ;-)
There are some cards which need to be jumpered as "second controller"
despite the fact that it's the only adapter. But these cards don't work
with the driver compiled into the kernel either.
> Gerard's) and various aic7xxx controllers. Others I'm not sure about;
> the Advansys driver *might* be exhibiting the same problem, but the
> bug reports aren't clear in this case.
>
> I get the impression that with the aic7xxx, it at least sometimes
> works as a module, but I've heard at least one report from someone
> else that one card (I think a 3940) which did not work with a module
> did work with a compiled-in driver. There may be issues like the
> amount of RAM involved, but I haven't seen a pattern yet.
Yes, most of the time the aic7xxx works as a module. I do not think that
the amount of RAM does matter. Most of the machines the problems were
reported had _lots_ of memory.
> It shows up a little more spectacularly with Red Hat Linux 4.0
> because we use initrd and SCSI modules to have a single boot
The same ist true for us. We do additionally ship disk images containing
"monolithic" kernels, so maybe there are cases where people do not report
problems because using another bootdisk everything works fine.
> image for everyone. Unfortunately, it took a while to track
> down, because at least some of these drivers appeared to work
> for some people and not others, and we're still working on getting
> more information. But I think we've got enough information at this
> point to indicate that this isn't a red herring.
>
> Any guesses as to the cause?
>
> michaelkjohnson
>
> "Ever wonder why the SAME PEOPLE make up ALL the conspiracy theories?"
Hubert mantel@suse.de