J. Ryan Earl wrote:

Brad Campbell wrote:

quoted

Just a heads up. I'm experiencing "issues" with 2.6.9-rc1 that don't 
occur with 2.6.5.

I run a 10 disk raid-5 on 3xPromise SATA150 cards and a 2 disk raid-0 
on the on-board VIA chipset.
The 10 disks are Maxtor Maxline-II SATA drives and the raid-0 is a 
pair of WD2000JB drives with
Addonics SATA->PATA converters. 

Did you try RAID-0 with the pair of Western Digital drives on the two 
free Promise slots?  If that fails similarly, it might not be a libata 
issue.

I thought about it. Given the drives are giving write errors when I write to /dev/md1 and not giving 
write errors when the raid is stopped and I write to the individual disks then it *must* be some 
issue with the hardware drivers.

If the block layer can submit a block to the driver that causes the drive to error out, then it's 
really a fault of the driver for trying to do something illegal.

Anyway, I have done a clone of the libata-2.6 bk tree and I'm trying to figure out how to extract 
all the individual csets between 2.6.6 and 2.6.7-rc1 so I can back them out 1 by 1 and see what 
caused the problem.

My other tack is to enable full SCSI debugging and compare the trace from 2.6.5 and 2.6.7-rc1 to see 
what is different.

I will give your suggestion a try when I get home tonight and fire it up on the extra promise channels.

Can anyone point me to a dummys guide to regression testing with BK? I read a great one by the WINE 
guys years ago on regression testing by date with CVS and it has proved immensely helpful over the 
years.

Regards,
Brad