x86_64 + 2.6.1-rc3 panics on aic79xx

x86_64 + 2.6.1-rc3 panics on aic79xx

[Berkley Shands]

	The default scsi driver for the adaptec 39320 (aic79xx) v 1.3.11
STILL causes panics and BUG halts when a drive decides to have troubles.
I have a 2.6.1-rc3 X86_64 kernel, with
ftp://ftp.x86-64.org/pub/linux/v2.6/x86_64-2.6.1rc2-1.bz2

applied. On the first scsi error, the kernel bughalts "BRKADRINT".
The 2.0.5 version of the aic79xx driver will not boot with the
x86-64 patch, regardless of the iommu=[off | nomerge | noforce] setting.

If I use a vanilla 2.6.1-rc3 kernel, and apply the 2.0.5 driver,
the system will not boot, stalling with a scsi0: Unexpected SEQINTCODE 26 (bad scb).
If I disable the iommu, (iommu=off), then the kernel will boot, but the
system does not work, as the TG3 Gig-E driver is 32 bit.

When the 2.6.1-rc3 + x86_64 patch ran, the I/O throughput on
raid0 was 50% of the 2.6.0 stock kernel (340-360 MB/Sec vs 670-880 MB/Sec).
The ioctl to change /dev/md0 read ahead parameter is being called.
There were some changes to the read ahead values (at least raid5) in
-rc3. Ouch!

	A pure stock 2.6.1-rc3 kernel also bug halts with sg corruption
on the first scsi retry. It seems that the iommu has serious issues
with all three variants of  current kernel patches.

berkley

Re: x86_64 + 2.6.1-rc3 panics on aic79xx

[James Bottomley]

On Thu, 2004-01-08 at 11:27, Berkley Shands wrote:
> 	A pure stock 2.6.1-rc3 kernel also bug halts with sg corruption
> on the first scsi retry. It seems that the iommu has serious issues
> with all three variants of  current kernel patches.

Andi Kleen has produced a patch for the sg list problems.  Could you see
if you can reproduce the issues when Andi produces a patch kit for
x86_64 and -rc3, so we're not trying to debug non-scsi issues?

Thanks,

James