Re: MORE: 2.0.33 SCSI arrrggghhh!

Doug Ledford (
Mon, 22 Dec 1997 17:52:42 -0600 (CST)

On 22-Dec-97 Dr. Werner Fink wrote:
>> 3. Here's the kernel configuration; from make config
>> Adaptec AIC7xxx support (CONFIG_SCSI_AIC7XXX) [Y/m/n/?]
>> Enable tagged command queueing (CONFIG_AIC7XXX_TAGGED_QUEUEING)
>> Override driver defaults for commands per LUN (CONFIG_OVERRIDE_CMDS)
>> [Y/n/?]
>> Maximum number of commands per LUN (CONFIG_AIC7XXX_CMDS_PER_LUN) [32]
>> Enable SCB paging (CONFIG_AIC7XXX_PAGE_ENABLE) [Y/n/?]
>Set CONFIG_AIC7XXX_PAGE_ENABLE to n and maybe the problems disapear.
>.. something goes wrong on heavy I/O with scb paging.

Actually, it's not SCB paging that causes problems. It's more like certain
devices do odd things with their SCSI commands that only show up when the
commands are being paged in and out. It's really a state issue. However,
with the 2.0.33 kernel code, it really shouldn't be a problem. After
reviewing the logs he had in his email, the problem appears to be something
else entirely and I sent him a small patch to test out. If it works
then the system should properly recover from the bus hang and go on with
life without any problems/corruption/etc. FWIW, with lots of drives and
tagged queueing enabled, the SCB paging code can make as much as a 20 to 30%
performance difference, so there is ample reason to make sure it works so
that the performance benefit can be achieved :)

E-Mail: Doug Ledford <>
Date: 22-Dec-97
Time: 17:52:44