AHA-2940UW problems with Linux driver
Francis Kubala
fkubala at bbn.com
Mon Aug 25 13:50:16 PDT 1997
I've stumbled onto a subtle and devastating problem that may involve Linux
and the Adaptec 2940UW while trying to bring up an external RAID system
controlled by CMD Viper IIs. The host was a P6 running Redhat Linux 2.0.30.
I was stress testing the RAID by inducing a single disk failure while under
load from the host. Almost every rebuild failed with several seemingly
random and unrelated symptoms such as RAID drives going offline and CMD
controller kernel panics. After swapping out and retesting every component
in the RAID system (with much consultation from the RAID integrator and CMD)
I swapped the 2940 on the host for a BusLogic 958. Since then, every test
has succeeded.
I have successfully used the 2940UWs on 30 P6 compute servers that are
running around the clock. They each have a single 10K RPM 9 GB Cheetah
system disk. The reliability of these systems fooled me into overlooking
the Adaptec for a long time. I am dumbstruck that a driver problem on the
host could cause a device to fail catastrophically on the other side of a
SCSI interconnect. This suggests to me that CMD may also have a problem.
But I have to believe that the driver/adapter combination is the source of
the problem.
CMD has indicated an interest in reproducing and analysing this problem in
their lab. I'm sure it would be invaluable to them (and to Linux) to have
members of the aic7xxx Linux team in the loop. If any of you are interested,
I'd like to put you in touch with them. I think Linux would benefit greatly
by having a rock-solid driver for this most popular adapter.
More information about the aic7xxx
mailing list