[563] in linux-scsi channel archive
Re: Errors & Unit Attention
daemon@ATHENA.MIT.EDU (Harik A'ttar)
Wed Aug 30 13:14:17 1995
Date: Wed, 30 Aug 1995 04:11:41 -0400 (EDT)
From: "Harik A'ttar" <harik@chaos.sub.ucf.edu>
Reply-To: drm89121@pegasus.cc.ucf.edu
To: Michael Weller <eowmob@exp-math.uni-essen.de>
cc: linux-scsi@vger.rutgers.edu
In-Reply-To: <Pine.3.89.9508291024.D93683-0100000@werner.exp-math.uni-essen.de>
On Tue, 29 Aug 1995, Michael Weller wrote:
>
> > From: sdw@lig.net (Stephen D. Williams)
> > Subject: Re: Errors & Unit Attention
> > To: bsa@kf8nh.wariat.org (Brandon S. Allbery KF8NH)
> > Date: Mon, 28 Aug 1995 10:22:26 -0400 (EDT)
> > Cc: sdw@lig.net, eric@aib.com, lnz@dandelion.com, linux-scsi@vger.rutgers.edu
>
> > > acquaintance also fail rather poorly in such a situation. (And for the
> > > multi-architecture folks: you can destroy an SBus SCSI card by trying to
> > > hot-plug devices...)
>
> Well hot swapping SCSI is indeed often done, but this is just dancing
> with the devil.. it usually works, but you never know. I never saw it as
> defined in the standard. (I might have overread it though, but I doubt it).
> Instead for whatever scsi-equipment you buy (even for expensive workstation
> scsi controllers of about a few thousand $) (and actually esp. for those
> as these guys do care more about their customers than vendors of PC cards)
> it explicitly warns you to NOT hot swap.
True. It is dangerous, but sometimes VERY useful (I.E. External
drives: put a normal drive in an external case, take it with you and
plug it into any machine you are at. It is nice not to have to power
down in these cases. :)
>
> Two mayor problems with hot swap exists: One is protocol related: If some
> traffic is on the scsi bus during the hot swap is is certainly interrupted
> what may not be detected in all cases, thus data corruption or device
> crashes are possible.
Well, if done right, you should be able to get the drive in correctly,
see below.
>
> The other problems are even worse: Static electricity may discharge
> during the hot plug and damage your equipment.
This is allways known as an Oops error. :)
>
> I know not only cases of instantly crashed unix work stations due to a hot
> plug but also mysterious dies of scsi-devices (usually adapters, not
> external devices) that might easily be related to hot plugging shortly
> before the device dead.
Again, it should be possible: just not with the normal: device to scsi
bus direct hardwired connection. That is a bad thing.
A much better way would be to actually use a 2 stage adapter: one part
plugs into the scsi bus and the power supply for the bus, and has an easy
way to ground before attaching it to the bus. Since it is a dead device
at start time, no data errors when you attach it. Powering up via the
devices normal power (arn't all of them those 4 plug thingys? Excepting
the 3.5 floppys.) The other piece plugs into the devices scsi connection,
and the devices power input. When the 2 are attached, the device powers
up (wait till the drive spins up) Then hit a reset button.
Of course, everything but the power cable is optically isolated, or else
you would have major troubles. However, all this really requires is
the Isolators, and have power cut to them all at the same time (Until
the reset has been performed.)
>
> From my point of view hot plugging scsi is one of the last tails in this
> world. Just don't do it. Even when common OS's (I mean unix
> implementations here, not DOS or something similar weird) ALLOW you to
> add/delete scsi devices on the fly.
>
> This is not to be confused with raid systems which use special controller
> hardware to allow hot plugging (of the raid disks).
True, but with just one connector similar to the above for each device
you want swapped (remember, the part on the bus can be removed once it goes
dead) You can safely install/remove the devices.
Of course, who wants to be the one testing this? :)
Put in the drive, drive is registered. Remove the drive. Drive is not
there. Put in the drive....
>
> Actually one of the advantage of most of these new serial optical busses to
> connect storage devices is that these are explicitly hot pluggable.
Sweet. However, optical busses are just a bit more expensive. Granted,
I would still love to be using one.
>
> Michael.
>
> (eowmob@exp-math.uni-essen.de or eowmob@pollux.exp-math.uni-essen.de
> Please do not use my vm or de0hrz1a accounts anymore. In case of real
> problems reaching me try mat42b@aixrs1.hrz.uni-essen.de instead.)
>
>
chaos@dynamic.ip.don't.reply Guess what? I really _DO_ speak for my
Dan Merillat / Harik A'ttar system. And if you share my opinions,
drm89121@pegasus.cc.ucf.edu you should seek professional help.
Finger for my PGP Public Key PGP-mail prefered, thank you.
