On Tue, 17 Sep 1996, Randy Gobbel wrote:
> I went back through recent kernel patches and found all the aic7xxx changes in
> 2.0.13. Sure, enough, backing out those changes (2.0.20 -> 2.0.12) restores
> the previous behavior. As in 2.0.20, I get a "queue full" message
It has been reported to some concurrent list that Quantum Atlas Wide
4.3GB and IBM DORS 2GB may return QUEUE FULL, even when only fiew tagged
commands are queued (4). Sunch drives can accept up to 16 queued commands.
It seems that disabling Tagged Command Queueing fixes the problem.
However for the Atlas with its 2MB cache it is a pity.
This behaviour seems to conforms with scsi-2 6.8.2:
"Tagged queuing may also be temporarily disabled internal
to the SCSI device during certain initialization periods or
to control internal resource utilization. During these periods the
target may elect to return QUEUE FULL status or it may respond to
any queue tag message with a MESSAGE REJECT"
Some SCSI drivers and subsystems are not ready to handle that properly.
I have prepared a patch for the ncr8xx driver that tries to handle such
unexpected QUEUE FULL status, but I cannot test it with my hardware.
Obviously, I cannot help you (unless you will decide to switch to a NCR
controller), but people who own an Atlas 4.3 GB or a DORS 2 GB can help
me to test the patch I have prepared for the 8xx driver.
Gerard.