Re: 2.4.20 st + aic7xxx (Adaptec 19160B) + VIA KT333 repeatable freeze

Kevin Easton (kevin@sylandro.com)
Mon, 16 Dec 2002 16:04:30 +1100


On Sun, Dec 15, 2002 at 01:01:45PM -0700, Justin T. Gibbs wrote:
> > On Fri, 13 Dec 2002 11:51:27 +1100,
> > Kevin Easton <kevin@sylandro.com> wrote:
> >> I'm not sure exactly where this problem fits in, but I'm getting a
> >> completely repeatable freeze (100% lockup, no response to keyboard)
> >> triggered by writing to /dev/st0 (dd if=/dev/urandom of=/dev/st0 bs=512
> >> count=163840 will reproduce it).
> >> So... does anyone have any ideas how I should start trying to track this
> >> down?
>
> You might also look into your BIOS to ensure that the option "PCI Byte
> Merging" is disabled. This option allows the chipset to perform illegal
> byte merging on the PCI bus that will upset the Adaptec. Since the byte
> merging will only occur in certain scenarios (heavily dependent on what
> is going on with the SCSI bus), you may only see the lockup when accessing
> a particular device or running a certain program.
>
> The latest versions of the aic7xxx and aic79xx drivers will automatically
> detect this broken VIA behavior and will fall back to using PIO for register
> access. Although I haven't generated patches against 2.4.20, you can pull
> down a src tarball for 2.4.X that should just drop in:

Well, like I thought, this BIOS (it's an ASUS A7V333 mobo) doesn't have
an option for PCI Byte Merging - it has very few PCI options at all.
I'm going to update the BIOS to see if that gains me anything useful.

I took a clean 2.4.20 source tree and overlaid it with the contents of
aic79xx-linux-2.4.20-20021213.tar.gz, and it doesn't seem to have fixed
the problem completely, though it is *much* harder to trigger it now - I
thought it was working, but the same problem reappeared after copying
just over 4.5Gb to a tape.

As I mentioned in my reply to Keith, the NMI watchdog isn't even
triggering after this crash, which seems a little odd to me.

- Kevin.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/