We're using AlphaPC164 boards with an Intraserver SCSI controller
that uses the NCR 53c875 code.

With two successive Seagate 2.1 GB ultrawide drives, we've seen
errors like this:

sd0(ncr0:0:0): M_REJECT received (0:8).
sd0(ncr0:0:0): M_REJECT received (0:8).
ncr0:0: ERROR (0:41) (9-a9-0) (f/3d) @ (bb8:19000000).
        script cmd = 88080000
	reg:     ca 10 80 3d 47 0f 00 0f 01 09 80 a9 80 01 19 08.
ncr0: have to clear fifos.
ncr0: restart (fatal error).
sd0(ncr0:0:0): COMMAND FAILED (9 ff) @0xfffffe004a5a8c00.
sd0(ncr0:0:0): COMMAND FAILED (9 ff) @0xfffffe004a5a9000.
sd0(ncr0:0:0): COMMAND FAILED (9 ff) @0xfffffe004a5a9400.
sd0(ncr0:0:0): COMMAND FAILED (9 2) @0xfffffe004a5a9800.
ncr0: timeout ccb=0xfffffe004a5a8c00 (skip)
ncr0: timeout ccb=0xfffffe004a5a9000 (skip)
ncr0: timeout ccb=0xfffffe004a5a9400 (skip)
ncr0: timeout ccb=0xfffffe004a5a9800 (skip)

after which the system is permanently hung.  Is there some reason
we can't restart a disk driver?  The problem can be repeated by
doing large file transfers, eg: cd x; tar cf - . | (cd y; tar xf -)
on something the size of the toolchain or kernel source code.  It
seems to vary in exactly when it happens as you vary the partition
table or newfs parameters (spindle speed moved it?!).

Anyone else have any history with this?  Any plans for error recovery
in the future?