[13518] in Athena Bugs
unaligned access error (AFS bug), decmips
daemon@ATHENA.MIT.EDU (jmmikkel@MIT.EDU)
Sat May 20 16:32:29 1995
From: jmmikkel@MIT.EDU
Date: Sat, 20 May 1995 16:32:28 -0400
To: bugs@MIT.EDU
In-Reply-To: [12721] in Athena Bugs
Last October I reported a strange bug which we guessed at the time may
have been AFS-caused, when on a dec, emacs suddenly wedged and the
xterm printed "pid 5137 (emacs) was killed on a kernel access at pc
0x430144" ad infinitum.
I got mail from yandros hypothesizing it was a "heisenbug" and until
recently had not run across it again.
I first saw it again when one of my 6.170 team members was running
antichess and I copied a new version of the executable from local disk
into AFS, after which the same thing happened to her, except the
error was probably "unaligned access" rather than "kernel
access". She was on a decstation (dragons-lair, I believe), and I was
not logged into the same machine, so the kernel didn't give a "Text
file busy"-type error (just so you don't have to ask me that).
Then someone mentioned a pid getting killed on an "unaligned access"
error on instance 6.170. For him, however, it only resulted in one
copy of the error message, not an infinite number. He said that it
happened kind of randomly when he ran his program. Nobody had changed
the file while he was using it. He was on a dec. When I suggested he
flush the file from the AFS cache, the problem went away.
I'm sort of guessing that it happens when the cache gets screwed up
wrt what's really stored on the server, but I just don't know enough
about AFS to guess any more...
Joanne
