SCSI-2 Parity Errors
Thomas.Clark at Eng.Sun.COM
Thu Mar 16 21:33:47 PST 1995
I agree with your comments about parity errors related to cabling
and bad devices. However, whether multiple devices are attached to
the same bus or there is a single device attached to the bus, SCSI
parity errors should be handled differently.
They are a cross-device problem in that the reporting of such an
error points to a potential problem in other devices attached to
the same bus.
If in fact the parity errors are encountered only on a single device,
deal with that device. Here I presume that all devices on the
bus support parity error checking. If not, then I have a problem
generating a response especially for those devices that do not
support parity checking, or those where parity checking has been
If the parity errors are reported against all attached devices, then
there is a bus problem and obviously something should be done. If
parity errors have been encountered on a subset of all attached
devices, then I have a different problem. Changes are I do not
have a device driver that can audit and correlate such errors
(which is again another problem).
In a multi-initiator environment the action taken should be the
result of a joint decision among initiators. Understand that this
"joint decision" is hard to implement, but "aborting and starting
over" is a rather unfriendly approach.
The appropriate response to a parity error just isn't simple to
determine. It has wide-ranging impacts that involve device drivers,
applications and networked systems.
Hopefully the Common Device Driver Working Group will address this
within a reasonable time. Hopefully solutions will be tailored to
handle the multi-initiator environment effectively and efficiently.
Hopefully device drivers will avoid pulling bus reset when a
parity error is encountered, or a target is reset because it
reported such an error. These are very touch to handle where there
are multiple initiators.
More information about the T10