About a month ago I mailed this list concerning a problem I
was experiencing with a KZPCM dual-port SCSI controller
and a test program I had written, which seemed to indicate
a possible controller problem.
The mail was picked up by IntraServer (who build the controller
and driver) who asked for details. They investigated and did
indeed find a problem. They reported back:
'Under the following conditions, data corruption may occur with the KZPCM:
1. An odd number of bytes must be written.
2. The disk being written to must be running in 'wide' mode.
3. The disk must be a Quantum Atlas III (this is what's inside a DEC
DS-RZ1EF-VW), as this drive (unlike all others they have tested)
performs a disconnect during the DATA OUT phase of a transfer.
Under these circumstances, the last byte of the data may fail to be written.'
They stress that under normal use by the operating system, transfers
are always padded to even bytes, and for the block device, transfers
are handled in 1K chunks so there should be no problem with the
KZPCM in normal use. I modified my test prog and confirm this.
They are working on a fix.
Cheers,
Terry.
Received on Mon Mar 29 1999 - 11:09:12 NZST