Observed on an i386/1.3 box, but sys/dev/pci/ncr* appear unchanged
between
1.3 and 1.3.3.

This box has a working Adaptec 2940U as ahc0, but I decided to move CDs
and DAT to another controller to reduce SCSI contention and bought a
cheaper Tekram 390U card (53c875 based, narrow SCSI bus).  

Everything seems to probe fine, eg:-
ncr0 at pci0 dev 9 function 0: ncr 53c875 fast20 wide scsi
ncr0: interrupting at irq 15
ncr0: minsync=12, maxsync=137, maxoffs=16, 128 dwords burst, large dma
fifo
ncr0: single-ended, open drain IRQ driver, using on-chip SRAM
ncr0: restart (scsi reset).
scsibus1 at ncr0: 16 targets
scsipi_inqmatch: 2/1/1 <, , >
st0 at scsibus1 targ 2 lun 0: <HP      HP35470A, HP35470A        , 7 09>
SCSI2 1
/sequential removable
st0: st0(ncr0:2:0): 5.0 MB/s (200 ns, offset 8)
drive empty
scsipi_inqmatch: 2/5/1 <, , >
scsipi_inqmatch: 2/5/1 <, , >
cd0 at scsibus1 targ 6 lun 0: <YAMAHA  CRW4260 , CRW4260         , 1.0g>
SCSI2 5
/cdrom removable

However accesses to scsibus1 (eg "mt -f /dev/st0 status") hang
indefinitely, with console messages like:-
ncr0: timeout ccb=0xf08ae800 (skip)

That seemed like it might be a case of dropped interrupts, so I tried
changing the irq mapped to that slot by the BIOS (it was originally
irq10, moved to 15 as the onboard IDE controllers are both disabled).

The system seemed stable for a while in this state (with a hung,
unkillable mt process), but left overnight scsibus0 on ahc0 seemed to
hang too and the box had to be reset.

Any ideas?  A scan of PRs and the post-i386 archive suggested these
cards mostly work except not enough CCBs are allocated if there more
than 4 drives (I tried upping that anyway to 256 --- no difference in
behaviour).

Pete.