ISC 2.2 Hangs on disk I/O

Ed Hall edhall at rand.org
Thu Mar 14 09:24:01 AEST 1991


In article <27DD4757.22595 at orion.oac.uci.edu> drector at orion.oac.uci.edu (David Rector) writes:
>In <24113 at hydra.gatech.EDU> doug at pravda (Doug MacKenzie) writes:
>>The problem:
>>  Some times during disk I/O, the disk light will just stay on solid
>>  and the system is locked up.  I can not get any response other than
>>  turning the stupid thing off.
>
>>  It happens sometimes during boot up when it is scanning the disk.
>>  It happens sometimes during large GREP's.
>>  It just seems to happen!!!
>>  This is not a panic.  It is just hung.
>
>Ah, another sufferer.  The disease seems to be incurable, but you won't
>notice it after a while.
>
>No one has firmly diagnosed this problem with the otherwise admirable
>WD1006, but the behavior suggests that an error condition is not being
>correctly handled by the ISC driver.  Conjecture: controller is asked
>to read a flakey sector; CRC comes up bad too many times and controller
>gives up; driver doesn't have a time-out and never sees error message
>or resets the controller; system hangs.

This could be, but I've also seen people on the net complaining about
this behavior with ESIX.  Both drivers may be broken--or the WD1006.

Years ago (well, maybe two) someone posted to comp.sys.ibm.pc a letter
from someone at Western Digital asking for people experiencing this
problem to contact her.  The implication was that they were going to
give up on solving the problem unless they found evidence that more
than a few individuals were affected.  I had just got my WD1006SV2 and
it hadn't frozen up, so like a dummy I didn't save the posting.  Of
course, within the next week or so my system froze.  But it's only
happened every few weeks, so I'm not too upset; if it were more than
just my home system I'd certainly have pursued is...

BTW, this is NOT the problem a batch of WD1006's had with overlapped
seeks.  That particular problem only existed on ISC 2.* systems with
two disks, and could be fixed by a small change to the driver's
configuration.

		-Ed Hall
		edhall at rand.org



More information about the Comp.unix.sysv386 mailing list