>> Are you sure?  I've [hot-plugged SCSI devices in and out] many times
>> without a problem.

> I've seen the SCSI port wedge on:
> Sun 3 hardware
> Sun sparc hardware
> Intel boxes with any of the Adaptec 154x and 174x family of SCSI controllers
> NCR UNIX boxes

> I can't say that I've ever tried it with NetBSD.  I've seen both SCO
> UNIX and SunOS 4.1.3 contaminate filesystems after SCSI termination
> was removed by a scavenger.  Solaris 2.x and others just wedge.

There's an important point here.  I have never tried completely hot
addition or removal of a SCSI device.  What I have done, often, is to
halt the cpu (BREAK, L1-A, whatever) to ensure signal stability, _then_
add or remove the device, then continue.  Provided I'm quick enough,
I've never had a problem (if I leave the cpu halted for long enough -
which means like minutes - I have seen the kernel crash or hang upon
continuing).

Note that I always make sure the bus is properly daisy-chained and
terminated before "continue"ing.  I also have been reasonably careful
about grounding, making sure that all boxes have a ground path before
frobbing the cabling.

I've been very careful to never do this when scsi transfers might take
place.  On a Sun, this means halting the cpu; on our Auspex, it means
ax_{add,remove}_device...I would certainly expect it to come to grief
if the chain is broken when a transfer is attempted.

					der Mouse

			    mouse@collatz.mcrcim.mcgill.edu